├── .gitignore
├── .gitmodules
├── .idea
    ├── misc.xml
    ├── modules.xml
    ├── other.xml
    ├── pseudo-3d.iml
    ├── vcs.xml
    └── workspace.xml
├── Dataset.py
├── README.md
├── data
    ├── .DS_Store
    ├── breakfastTrainTestList
    │   ├── classInd.txt
    │   ├── fixclass.py
    │   ├── testlist01.txt
    │   ├── trainlist01.txt
    │   └── validationlist01.txt
    ├── extract.py
    ├── makeVideoFolder.py
    ├── merlTrainTestList
    │   ├── classInd.txt
    │   ├── merl.py
    │   ├── testlist01.txt
    │   ├── trainlist01.txt
    │   └── validationlist01.txt
    ├── movefile.py
    └── ucfTrainTestlist
    │   ├── .DS_Store
    │   ├── classInd.txt
    │   ├── testlist01.txt
    │   ├── testlist02.txt
    │   ├── testlist03.txt
    │   ├── trainlist01.txt
    │   ├── trainlist02.txt
    │   ├── trainlist03.txt
    │   ├── validationlist01.txt
    │   ├── validationlist02.txt
    │   └── validationlist03.txt
├── dog.0.jpg
├── logger.py
├── lr_scheduler.py
├── main.py
├── meter.py
├── models
    ├── C3D.py
    ├── i3d.py
    ├── i3dpt.py
    └── p3d_model.py
├── test.py
├── train.py
├── transforms.py
├── utils.py
├── video_transforms.py
└── visualize.py


/.gitignore:
--------------------------------------------------------------------------------
1 | *.tar
2 | .DS_Store
3 | __pycache__
4 | lib/
5 | Breakfast/
6 | UCF101/
7 | *.pickle
8 | *.pth
9 | *.tar


--------------------------------------------------------------------------------
/.gitmodules:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/naviocean/pseudo-3d-pytorch/50297d11248630792709782f467982e80c281384/.gitmodules


--------------------------------------------------------------------------------
/.idea/misc.xml:
--------------------------------------------------------------------------------
1 | <?xml version="1.0" encoding="UTF-8"?>
2 | <project version="4">
3 |   <component name="JavaScriptSettings">
4 |     <option name="languageLevel" value="ES6" />
5 |   </component>
6 |   <component name="ProjectRootManager" version="2" project-jdk-name="Python 3.6 (myconda)" project-jdk-type="Python SDK" />
7 | </project>


--------------------------------------------------------------------------------
/.idea/modules.xml:
--------------------------------------------------------------------------------
1 | <?xml version="1.0" encoding="UTF-8"?>
2 | <project version="4">
3 |   <component name="ProjectModuleManager">
4 |     <modules>
5 |       <module fileurl="file://$PROJECT_DIR$/.idea/pseudo-3d.iml" filepath="$PROJECT_DIR$/.idea/pseudo-3d.iml" />
6 |     </modules>
7 |   </component>
8 | </project>


--------------------------------------------------------------------------------
/.idea/other.xml:
--------------------------------------------------------------------------------
1 | <?xml version="1.0" encoding="UTF-8"?>
2 | <project version="4">
3 |   <component name="PySciProjectComponent">
4 |     <option name="PY_SCI_VIEW_SUGGESTED" value="true" />
5 |   </component>
6 | </project>


--------------------------------------------------------------------------------
/.idea/pseudo-3d.iml:
--------------------------------------------------------------------------------
 1 | <?xml version="1.0" encoding="UTF-8"?>
 2 | <module type="PYTHON_MODULE" version="4">
 3 |   <component name="NewModuleRootManager">
 4 |     <content url="file://$MODULE_DIR$" />
 5 |     <orderEntry type="jdk" jdkName="Python 3.6 (myconda)" jdkType="Python SDK" />
 6 |     <orderEntry type="sourceFolder" forTests="false" />
 7 |   </component>
 8 |   <component name="TestRunnerService">
 9 |     <option name="PROJECT_TEST_RUNNER" value="Unittests" />
10 |   </component>
11 | </module>


--------------------------------------------------------------------------------
/.idea/vcs.xml:
--------------------------------------------------------------------------------
1 | <?xml version="1.0" encoding="UTF-8"?>
2 | <project version="4">
3 |   <component name="VcsDirectoryMappings">
4 |     <mapping directory="$PROJECT_DIR$" vcs="Git" />
5 |   </component>
6 | </project>


--------------------------------------------------------------------------------
/.idea/workspace.xml:
--------------------------------------------------------------------------------
  1 | <?xml version="1.0" encoding="UTF-8"?>
  2 | <project version="4">
  3 |   <component name="ChangeListManager">
  4 |     <list default="true" id="479e4968-bd6f-4bff-80fd-a0aeab5d9edf" name="Default Changelist" comment="``">
  5 |       <change beforePath="$PROJECT_DIR$/train.py" beforeDir="false" afterPath="$PROJECT_DIR$/train.py" afterDir="false" />
  6 |     </list>
  7 |     <option name="EXCLUDED_CONVERTED_TO_IGNORED" value="true" />
  8 |     <option name="SHOW_DIALOG" value="false" />
  9 |     <option name="HIGHLIGHT_CONFLICTS" value="true" />
 10 |     <option name="HIGHLIGHT_NON_ACTIVE_CHANGELIST" value="false" />
 11 |     <option name="LAST_RESOLUTION" value="IGNORE" />
 12 |   </component>
 13 |   <component name="FUSProjectUsageTrigger">
 14 |     <session id="-2110775552">
 15 |       <usages-collector id="statistics.lifecycle.project">
 16 |         <counts>
 17 |           <entry key="project.closed" value="27" />
 18 |           <entry key="project.open.time.0" value="2" />
 19 |           <entry key="project.open.time.1" value="4" />
 20 |           <entry key="project.open.time.4" value="2" />
 21 |           <entry key="project.open.time.5" value="11" />
 22 |           <entry key="project.open.time.6" value="14" />
 23 |           <entry key="project.open.time.7" value="6" />
 24 |           <entry key="project.open.time.8" value="3" />
 25 |           <entry key="project.opened" value="42" />
 26 |         </counts>
 27 |       </usages-collector>
 28 |       <usages-collector id="statistics.file.extensions.open">
 29 |         <counts>
 30 |           <entry key="gitignore" value="10" />
 31 |           <entry key="gitmodules" value="7" />
 32 |           <entry key="jpg" value="2" />
 33 |           <entry key="log" value="5" />
 34 |           <entry key="md" value="4" />
 35 |           <entry key="py" value="108" />
 36 |           <entry key="txt" value="8" />
 37 |           <entry key="vmoptions" value="1" />
 38 |         </counts>
 39 |       </usages-collector>
 40 |       <usages-collector id="statistics.file.types.open">
 41 |         <counts>
 42 |           <entry key="Image" value="2" />
 43 |           <entry key="Log" value="5" />
 44 |           <entry key="Markdown" value="4" />
 45 |           <entry key="PLAIN_TEXT" value="26" />
 46 |           <entry key="Python" value="106" />
 47 |           <entry key="Scratch" value="2" />
 48 |         </counts>
 49 |       </usages-collector>
 50 |       <usages-collector id="statistics.file.extensions.edit">
 51 |         <counts>
 52 |           <entry key="Python Console" value="14" />
 53 |           <entry key="dummy" value="20" />
 54 |           <entry key="gitignore" value="69" />
 55 |           <entry key="gitmodules" value="13" />
 56 |           <entry key="md" value="901" />
 57 |           <entry key="py" value="15368" />
 58 |           <entry key="txt" value="1" />
 59 |           <entry key="vmoptions" value="4" />
 60 |         </counts>
 61 |       </usages-collector>
 62 |       <usages-collector id="statistics.file.types.edit">
 63 |         <counts>
 64 |           <entry key="Markdown" value="901" />
 65 |           <entry key="PLAIN_TEXT" value="107" />
 66 |           <entry key="Python" value="15361" />
 67 |           <entry key="Scratch" value="21" />
 68 |         </counts>
 69 |       </usages-collector>
 70 |     </session>
 71 |   </component>
 72 |   <component name="FileEditorManager">
 73 |     <leaf SIDE_TABS_SIZE_LIMIT_KEY="300">
 74 |       <file pinned="false" current-in-tab="true">
 75 |         <entry file="file://$PROJECT_DIR$/train.py">
 76 |           <provider selected="true" editor-type-id="text-editor">
 77 |             <state relative-caret-position="792">
 78 |               <caret line="381" column="8" selection-start-line="381" selection-start-column="8" selection-end-line="381" selection-end-column="8" />
 79 |               <folding>
 80 |                 <element signature="e#0#37#0" expanded="true" />
 81 |                 <marker date="1536296846868" expanded="true" signature="7009:7014" ph="..." />
 82 |               </folding>
 83 |             </state>
 84 |           </provider>
 85 |         </entry>
 86 |       </file>
 87 |       <file pinned="false" current-in-tab="false">
 88 |         <entry file="file://$PROJECT_DIR$/visualize.py">
 89 |           <provider selected="true" editor-type-id="text-editor">
 90 |             <state relative-caret-position="195">
 91 |               <caret line="13" column="41" selection-start-line="13" selection-start-column="37" selection-end-line="13" selection-end-column="41" />
 92 |             </state>
 93 |           </provider>
 94 |         </entry>
 95 |       </file>
 96 |       <file pinned="false" current-in-tab="false">
 97 |         <entry file="file://$PROJECT_DIR$/models/p3d_model.py">
 98 |           <provider selected="true" editor-type-id="text-editor">
 99 |             <state relative-caret-position="807">
100 |               <caret line="388" column="47" lean-forward="true" selection-start-line="387" selection-start-column="4" selection-end-line="388" selection-end-column="47" />
101 |             </state>
102 |           </provider>
103 |         </entry>
104 |       </file>
105 |       <file pinned="false" current-in-tab="false">
106 |         <entry file="file://$PROJECT_DIR$/test.py">
107 |           <provider selected="true" editor-type-id="text-editor">
108 |             <state relative-caret-position="1530">
109 |               <caret line="120" column="38" selection-start-line="120" selection-start-column="23" selection-end-line="120" selection-end-column="38" />
110 |               <folding>
111 |                 <element signature="e#0#37#0" expanded="true" />
112 |               </folding>
113 |             </state>
114 |           </provider>
115 |         </entry>
116 |       </file>
117 |       <file pinned="false" current-in-tab="false">
118 |         <entry file="file://$PROJECT_DIR$/.gitignore">
119 |           <provider selected="true" editor-type-id="text-editor">
120 |             <state relative-caret-position="120">
121 |               <caret line="8" column="5" selection-start-line="8" selection-start-column="5" selection-end-line="8" selection-end-column="5" />
122 |             </state>
123 |           </provider>
124 |         </entry>
125 |       </file>
126 |       <file pinned="false" current-in-tab="false">
127 |         <entry file="file://$PROJECT_DIR$/Dataset.py">
128 |           <provider selected="true" editor-type-id="text-editor">
129 |             <state relative-caret-position="495">
130 |               <caret line="38" column="19" selection-start-line="38" selection-start-column="19" selection-end-line="38" selection-end-column="19" />
131 |               <folding>
132 |                 <element signature="e#0#21#0" expanded="true" />
133 |               </folding>
134 |             </state>
135 |           </provider>
136 |         </entry>
137 |       </file>
138 |       <file pinned="false" current-in-tab="false">
139 |         <entry file="file://$PROJECT_DIR$/main.py">
140 |           <provider selected="true" editor-type-id="text-editor">
141 |             <state relative-caret-position="510">
142 |               <caret line="37" column="44" lean-forward="true" selection-start-line="37" selection-start-column="44" selection-end-line="37" selection-end-column="44" />
143 |             </state>
144 |           </provider>
145 |         </entry>
146 |       </file>
147 |     </leaf>
148 |   </component>
149 |   <component name="FileTemplateManagerImpl">
150 |     <option name="RECENT_TEMPLATES">
151 |       <list>
152 |         <option value="Python Script" />
153 |       </list>
154 |     </option>
155 |   </component>
156 |   <component name="FindInProjectRecents">
157 |     <findStrings>
158 |       <find>num_frames</find>
159 |       <find>frames</find>
160 |       <find>RandomResizedCrop</find>
161 |       <find>name_list</find>
162 |       <find>C3D</find>
163 |       <find>clip = self.get_video_tensor(frames_path)</find>
164 |       <find>validate</find>
165 |       <find>data =</find>
166 |       <find>normalize</find>
167 |       <find>Stack</find>
168 |       <find>roll</find>
169 |       <find>mean</find>
170 |       <find>momentum</find>
171 |       <find>np.</find>
172 |       <find>data</find>
173 |       <find>160</find>
174 |       <find>Nor</find>
175 |       <find>Normalize</find>
176 |       <find>self.data_folder</find>
177 |       <find>criterion</find>
178 |       <find>216</find>
179 |       <find>get_optim_policies</find>
180 |       <find>16</find>
181 |       <find>pretrained</find>
182 |       <find>check_gpu</find>
183 |       <find>transfer_model</find>
184 |       <find>trans</find>
185 |       <find>self.ac</find>
186 |       <find>transfer</find>
187 |       <find>RandomSizedCrop</find>
188 |     </findStrings>
189 |     <replaceStrings>
190 |       <replace>check_gpu()</replace>
191 |     </replaceStrings>
192 |     <dirStrings>
193 |       <dir>$PROJECT_DIR$/videotransforms</dir>
194 |       <dir>$PROJECT_DIR$</dir>
195 |     </dirStrings>
196 |   </component>
197 |   <component name="Git.Settings">
198 |     <option name="RECENT_GIT_ROOT_PATH" value="$PROJECT_DIR$" />
199 |   </component>
200 |   <component name="IdeDocumentHistory">
201 |     <option name="CHANGED_PATHS">
202 |       <list>
203 |         <option value="$PROJECT_DIR$/AverageMeter.py" />
204 |         <option value="$PROJECT_DIR$/train2.py" />
205 |         <option value="$PROJECT_DIR$/logger.py" />
206 |         <option value="$PROJECT_DIR$/README.md" />
207 |         <option value="$PROJECT_DIR$/data/breakfastTrainTestList/fixclass.py" />
208 |         <option value="$APPLICATION_CONFIG_DIR$/scratches/scratch.py" />
209 |         <option value="$PROJECT_DIR$/ufc101_dataset.py" />
210 |         <option value="$PROJECT_DIR$/breakfirst_dataset.py" />
211 |         <option value="$PROJECT_DIR$/merl_dataset.py" />
212 |         <option value="$PROJECT_DIR$/lib/video_transforms/testransforms.py" />
213 |         <option value="$PROJECT_DIR$/lib/video_transforms/videotransforms/tensor_transforms.py" />
214 |         <option value="$PROJECT_DIR$/lib/video_transforms/videotransforms/volume_transforms.py" />
215 |         <option value="$PROJECT_DIR$/video_transforms.py" />
216 |         <option value="$APPLICATION_CONFIG_DIR$/pycharm.vmoptions" />
217 |         <option value="$PROJECT_DIR$/Dataset.py" />
218 |         <option value="$PROJECT_DIR$/C3D.py" />
219 |         <option value="$PROJECT_DIR$/i3dpt.py" />
220 |         <option value="$PROJECT_DIR$/i3d.py" />
221 |         <option value="$PROJECT_DIR$/.gitignore" />
222 |         <option value="$PROJECT_DIR$/lib/p3d_model.py" />
223 |         <option value="$PROJECT_DIR$/.gitmodules" />
224 |         <option value="$PROJECT_DIR$/utils.py" />
225 |         <option value="$PROJECT_DIR$/models/p3d_model.py" />
226 |         <option value="$PROJECT_DIR$/transforms.py" />
227 |         <option value="$PROJECT_DIR$/test.py" />
228 |         <option value="$PROJECT_DIR$/visualize.py" />
229 |         <option value="$PROJECT_DIR$/main.py" />
230 |         <option value="$PROJECT_DIR$/train.py" />
231 |       </list>
232 |     </option>
233 |   </component>
234 |   <component name="JsBuildToolGruntFileManager" detection-done="true" sorting="DEFINITION_ORDER" />
235 |   <component name="JsBuildToolPackageJson" detection-done="true" sorting="DEFINITION_ORDER" />
236 |   <component name="JsGulpfileManager">
237 |     <detection-done>true</detection-done>
238 |     <sorting>DEFINITION_ORDER</sorting>
239 |   </component>
240 |   <component name="ProjectFrameBounds">
241 |     <option name="x" value="972" />
242 |     <option name="y" value="-1393" />
243 |     <option name="width" value="1570" />
244 |     <option name="height" value="1338" />
245 |   </component>
246 |   <component name="ProjectInspectionProfilesVisibleTreeState">
247 |     <entry key="Project Default">
248 |       <profile-state>
249 |         <expanded-state>
250 |           <State />
251 |           <State>
252 |             <id>Python</id>
253 |           </State>
254 |         </expanded-state>
255 |         <selected-state>
256 |           <State>
257 |             <id>PyCompatibilityInspection</id>
258 |           </State>
259 |         </selected-state>
260 |       </profile-state>
261 |     </entry>
262 |   </component>
263 |   <component name="ProjectLevelVcsManager" settingsEditedManually="true" />
264 |   <component name="ProjectView">
265 |     <navigator proportions="" version="1">
266 |       <foldersAlwaysOnTop value="true" />
267 |     </navigator>
268 |     <panes>
269 |       <pane id="Scope" />
270 |       <pane id="ProjectPane">
271 |         <subPane>
272 |           <expand>
273 |             <path>
274 |               <item name="pseudo-3d" type="b2602c69:ProjectViewProjectNode" />
275 |               <item name="pseudo-3d" type="462c0819:PsiDirectoryNode" />
276 |             </path>
277 |             <path>
278 |               <item name="pseudo-3d" type="b2602c69:ProjectViewProjectNode" />
279 |               <item name="pseudo-3d" type="462c0819:PsiDirectoryNode" />
280 |               <item name="models" type="462c0819:PsiDirectoryNode" />
281 |             </path>
282 |             <path>
283 |               <item name="pseudo-3d" type="b2602c69:ProjectViewProjectNode" />
284 |               <item name="External Libraries" type="cb654da1:ExternalLibrariesNode" />
285 |             </path>
286 |           </expand>
287 |           <select />
288 |         </subPane>
289 |       </pane>
290 |     </panes>
291 |   </component>
292 |   <component name="PropertiesComponent">
293 |     <property name="SearchEverywhereHistoryKey" value="&#9;FILE&#9;file:///Users/naviocean/Desktop/test/pseudo-3d/train.py" />
294 |     <property name="WebServerToolWindowFactoryState" value="false" />
295 |     <property name="com.intellij.ide.scratch.LRUPopupBuilder$1/New Scratch File" value="Python" />
296 |     <property name="last_opened_file_path" value="$PROJECT_DIR$" />
297 |     <property name="nodejs_interpreter_path.stuck_in_default_project" value="undefined stuck path" />
298 |     <property name="nodejs_npm_path_reset_for_default_project" value="true" />
299 |     <property name="settings.editor.selected.configurable" value="project.propVCSSupport.Mappings" />
300 |   </component>
301 |   <component name="RecentsManager">
302 |     <key name="CopyFile.RECENT_KEYS">
303 |       <recent name="$PROJECT_DIR$/models" />
304 |       <recent name="$PROJECT_DIR$" />
305 |       <recent name="$PROJECT_DIR$/data" />
306 |     </key>
307 |     <key name="MoveFile.RECENT_KEYS">
308 |       <recent name="$PROJECT_DIR$/models" />
309 |       <recent name="$PROJECT_DIR$" />
310 |     </key>
311 |   </component>
312 |   <component name="RunDashboard">
313 |     <option name="ruleStates">
314 |       <list>
315 |         <RuleState>
316 |           <option name="name" value="ConfigurationTypeDashboardGroupingRule" />
317 |         </RuleState>
318 |         <RuleState>
319 |           <option name="name" value="StatusDashboardGroupingRule" />
320 |         </RuleState>
321 |       </list>
322 |     </option>
323 |   </component>
324 |   <component name="SvnConfiguration">
325 |     <configuration />
326 |   </component>
327 |   <component name="TaskManager">
328 |     <task active="true" id="Default" summary="Default task">
329 |       <changelist id="479e4968-bd6f-4bff-80fd-a0aeab5d9edf" name="Default Changelist" comment="" />
330 |       <created>1534148020669</created>
331 |       <option name="number" value="Default" />
332 |       <option name="presentableId" value="Default" />
333 |       <updated>1534148020669</updated>
334 |     </task>
335 |     <servers />
336 |   </component>
337 |   <component name="TodoView">
338 |     <todo-panel id="selected-file">
339 |       <is-autoscroll-to-source value="true" />
340 |     </todo-panel>
341 |     <todo-panel id="all">
342 |       <are-packages-shown value="true" />
343 |       <is-autoscroll-to-source value="true" />
344 |     </todo-panel>
345 |   </component>
346 |   <component name="ToolWindowManager">
347 |     <frame x="972" y="-1393" width="1570" height="1338" extended-state="0" />
348 |     <editor active="true" />
349 |     <layout>
350 |       <window_info content_ui="combo" id="Project" order="0" visible="true" weight="0.49803665" />
351 |       <window_info id="Structure" order="1" side_tool="true" weight="0.25" />
352 |       <window_info id="Favorites" order="2" side_tool="true" />
353 |       <window_info anchor="bottom" id="Message" order="0" />
354 |       <window_info anchor="bottom" id="Find" order="1" weight="0.32985553" />
355 |       <window_info anchor="bottom" id="Run" order="2" />
356 |       <window_info anchor="bottom" id="Debug" order="3" weight="0.4" />
357 |       <window_info anchor="bottom" id="Cvs" order="4" weight="0.25" />
358 |       <window_info anchor="bottom" id="Inspection" order="5" weight="0.4" />
359 |       <window_info anchor="bottom" id="TODO" order="6" weight="0.32985553" />
360 |       <window_info anchor="bottom" id="Docker" order="7" show_stripe_button="false" />
361 |       <window_info anchor="bottom" id="Version Control" order="8" />
362 |       <window_info anchor="bottom" id="Database Changes" order="9" show_stripe_button="false" />
363 |       <window_info anchor="bottom" id="Event Log" order="10" sideWeight="0.5052356" side_tool="true" weight="0.2070626" />
364 |       <window_info active="true" anchor="bottom" id="Terminal" order="11" sideWeight="0.4947644" visible="true" weight="0.2070626" />
365 |       <window_info anchor="bottom" id="Python Console" order="12" weight="0.32981133" />
366 |       <window_info anchor="bottom" id="Messages" order="13" />
367 |       <window_info anchor="right" id="Commander" internal_type="SLIDING" order="0" type="SLIDING" weight="0.4" />
368 |       <window_info anchor="right" id="Ant Build" order="1" weight="0.25" />
369 |       <window_info anchor="right" content_ui="combo" id="Hierarchy" order="2" weight="0.25" />
370 |       <window_info anchor="right" id="SciView" order="3" />
371 |       <window_info anchor="right" id="Database" order="4" />
372 |     </layout>
373 |     <layout-to-restore>
374 |       <window_info content_ui="combo" id="Project" order="0" visible="true" weight="0.2629827" />
375 |       <window_info id="Structure" order="1" side_tool="true" weight="0.25" />
376 |       <window_info id="Favorites" order="2" side_tool="true" />
377 |       <window_info anchor="bottom" id="Messages" order="0" />
378 |       <window_info anchor="bottom" id="Message" order="1" />
379 |       <window_info anchor="bottom" id="Find" order="2" />
380 |       <window_info anchor="bottom" id="Run" order="3" />
381 |       <window_info anchor="bottom" id="Debug" order="4" weight="0.4" />
382 |       <window_info anchor="bottom" id="Cvs" order="5" weight="0.25" />
383 |       <window_info anchor="bottom" id="Inspection" order="6" weight="0.4" />
384 |       <window_info anchor="bottom" id="TODO" order="7" />
385 |       <window_info anchor="bottom" id="Docker" order="8" show_stripe_button="false" />
386 |       <window_info anchor="bottom" id="Version Control" order="9" />
387 |       <window_info anchor="bottom" id="Database Changes" order="10" show_stripe_button="false" />
388 |       <window_info anchor="bottom" id="Event Log" order="11" side_tool="true" />
389 |       <window_info active="true" anchor="bottom" id="Terminal" order="12" visible="true" weight="0.21687697" />
390 |       <window_info anchor="bottom" id="Python Console" order="13" weight="0.32981133" />
391 |       <window_info anchor="right" id="Commander" internal_type="SLIDING" order="0" type="SLIDING" weight="0.4" />
392 |       <window_info anchor="right" id="Ant Build" order="1" weight="0.25" />
393 |       <window_info anchor="right" content_ui="combo" id="Hierarchy" order="2" weight="0.25" />
394 |       <window_info anchor="right" id="SciView" order="3" />
395 |       <window_info anchor="right" id="Database" order="4" />
396 |     </layout-to-restore>
397 |   </component>
398 |   <component name="TypeScriptGeneratedFilesManager">
399 |     <option name="version" value="1" />
400 |   </component>
401 |   <component name="VcsContentAnnotationSettings">
402 |     <option name="myLimit" value="2678400000" />
403 |   </component>
404 |   <component name="VcsManagerConfiguration">
405 |     <ignored-roots>
406 |       <path value="$PROJECT_DIR$/lib" />
407 |     </ignored-roots>
408 |   </component>
409 |   <component name="editorHistoryManager">
410 |     <entry file="file://$PROJECT_DIR$/data/makeVideoFolder.py">
411 |       <provider selected="true" editor-type-id="text-editor">
412 |         <state relative-caret-position="495">
413 |           <caret line="33" column="15" lean-forward="true" selection-start-line="33" selection-start-column="15" selection-end-line="33" selection-end-column="15" />
414 |         </state>
415 |       </provider>
416 |     </entry>
417 |     <entry file="file://$PROJECT_DIR$/data/movefile.py">
418 |       <provider selected="true" editor-type-id="text-editor">
419 |         <state relative-caret-position="285">
420 |           <caret line="20" column="54" lean-forward="true" selection-start-line="20" selection-start-column="7" selection-end-line="20" selection-end-column="54" />
421 |         </state>
422 |       </provider>
423 |     </entry>
424 |     <entry file="file://$PROJECT_DIR$/data/ucfTrainTestlist/testlist01.txt">
425 |       <provider selected="true" editor-type-id="text-editor" />
426 |     </entry>
427 |     <entry file="file://$PROJECT_DIR$/data/breakfastTrainTestList/validationlist01.txt">
428 |       <provider selected="true" editor-type-id="text-editor" />
429 |     </entry>
430 |     <entry file="file://$PROJECT_DIR$/log.txt" />
431 |     <entry file="file://$PROJECT_DIR$/data/breakfastTrainTestList/testlist01.txt">
432 |       <provider selected="true" editor-type-id="text-editor" />
433 |     </entry>
434 |     <entry file="file://$PROJECT_DIR$/data/breakfastTrainTestList/fixclass.py">
435 |       <provider selected="true" editor-type-id="text-editor">
436 |         <state relative-caret-position="510">
437 |           <caret line="34" column="25" selection-start-line="34" selection-start-column="25" selection-end-line="34" selection-end-column="25" />
438 |         </state>
439 |       </provider>
440 |     </entry>
441 |     <entry file="file://$PROJECT_DIR$/data/breakfastTrainTestList/trainlist01.txt">
442 |       <provider selected="true" editor-type-id="text-editor">
443 |         <state relative-caret-position="784">
444 |           <caret line="336" selection-start-line="336" selection-end-line="336" />
445 |         </state>
446 |       </provider>
447 |     </entry>
448 |     <entry file="file://$PROJECT_DIR$/data/breakfastTrainTestList/classInd.txt">
449 |       <provider selected="true" editor-type-id="text-editor">
450 |         <state relative-caret-position="405">
451 |           <caret line="27" selection-start-line="27" selection-end-line="27" />
452 |         </state>
453 |       </provider>
454 |     </entry>
455 |     <entry file="file://$APPLICATION_CONFIG_DIR$/scratches/scratch.py" />
456 |     <entry file="file://$PROJECT_DIR$/UCF101/test.log" />
457 |     <entry file="file://$APPLICATION_CONFIG_DIR$/scratches/scratch_1.py" />
458 |     <entry file="file://$PROJECT_DIR$/Breakfast/test.log" />
459 |     <entry file="file://$PROJECT_DIR$/Breakfast/train.log" />
460 |     <entry file="file://$PROJECT_DIR$/UCF101/train.log" />
461 |     <entry file="file://$PROJECT_DIR$/merl_dataset.py" />
462 |     <entry file="file://$PROJECT_DIR$/breakfirst_dataset.py" />
463 |     <entry file="file://$PROJECT_DIR$/ufc101_dataset.py" />
464 |     <entry file="jar:///anaconda3/envs/myconda/lib/python3.6/site-packages/torchvision-0.2.1-py3.6.egg!/torchvision/transforms/transforms.py">
465 |       <provider selected="true" editor-type-id="text-editor">
466 |         <state relative-caret-position="258">
467 |           <caret line="480" column="6" selection-start-line="480" selection-start-column="6" selection-end-line="480" selection-end-column="6" />
468 |         </state>
469 |       </provider>
470 |     </entry>
471 |     <entry file="file://$PROJECT_DIR$/logger.py">
472 |       <provider selected="true" editor-type-id="text-editor">
473 |         <state relative-caret-position="240">
474 |           <caret line="16" selection-start-line="16" selection-end-line="16" />
475 |         </state>
476 |       </provider>
477 |     </entry>
478 |     <entry file="file:///anaconda3/envs/myconda/lib/python3.6/site-packages/torch/utils/data/__init__.py">
479 |       <provider selected="true" editor-type-id="text-editor" />
480 |     </entry>
481 |     <entry file="file:///anaconda3/envs/myconda/lib/python3.6/site-packages/torch/utils/data/dataloader.py">
482 |       <provider selected="true" editor-type-id="text-editor">
483 |         <state relative-caret-position="774">
484 |           <caret line="449" column="8" selection-start-line="449" selection-start-column="8" selection-end-line="449" selection-end-column="8" />
485 |         </state>
486 |       </provider>
487 |     </entry>
488 |     <entry file="file://$PROJECT_DIR$/lib/video_transforms/videotransforms/stack_transforms.py" />
489 |     <entry file="file://$PROJECT_DIR$/lib/video_transforms/videotransforms/functional.py" />
490 |     <entry file="file://$PROJECT_DIR$/lib/video_transforms/videotransforms/video_transforms.py" />
491 |     <entry file="file://$PROJECT_DIR$/lib/video_transforms/videotransforms/tensor_transforms.py" />
492 |     <entry file="file://$PROJECT_DIR$/lib/video_transforms/videotransforms/volume_transforms.py" />
493 |     <entry file="file://$PROJECT_DIR$/lib/video_transforms/testransforms.py" />
494 |     <entry file="file://$PROJECT_DIR$/videotransforms/stack_transforms.py" />
495 |     <entry file="file://$PROJECT_DIR$/videotransforms/functional.py" />
496 |     <entry file="file://$PROJECT_DIR$/videotransforms/volume_transforms.py" />
497 |     <entry file="file://$PROJECT_DIR$/videotransforms/video_transforms.py" />
498 |     <entry file="file://$PROJECT_DIR$/videotransforms/tensor_transforms.py" />
499 |     <entry file="file://$APPLICATION_CONFIG_DIR$/pycharm.vmoptions">
500 |       <provider selected="true" editor-type-id="text-editor">
501 |         <state relative-caret-position="240">
502 |           <caret line="16" column="55" selection-start-line="16" selection-start-column="55" selection-end-line="16" selection-end-column="55" />
503 |         </state>
504 |       </provider>
505 |     </entry>
506 |     <entry file="file://$PROJECT_DIR$/models/i3d.py">
507 |       <provider selected="true" editor-type-id="text-editor">
508 |         <state relative-caret-position="911">
509 |           <caret line="328" column="25" selection-start-line="328" selection-start-column="25" selection-end-line="328" selection-end-column="25" />
510 |         </state>
511 |       </provider>
512 |     </entry>
513 |     <entry file="file://$PROJECT_DIR$/models/i3dpt.py">
514 |       <provider selected="true" editor-type-id="text-editor">
515 |         <state relative-caret-position="5714">
516 |           <caret line="433" column="25" selection-start-line="433" selection-start-column="25" selection-end-line="433" selection-end-column="25" />
517 |         </state>
518 |       </provider>
519 |     </entry>
520 |     <entry file="file://$PROJECT_DIR$/models/C3D.py">
521 |       <provider selected="true" editor-type-id="text-editor">
522 |         <state relative-caret-position="818">
523 |           <caret line="69" selection-start-line="69" selection-end-line="74" selection-end-column="3" />
524 |         </state>
525 |       </provider>
526 |     </entry>
527 |     <entry file="file://$PROJECT_DIR$/video_transforms.py">
528 |       <provider selected="true" editor-type-id="text-editor">
529 |         <state relative-caret-position="911">
530 |           <caret line="113" column="54" selection-start-line="113" selection-start-column="54" selection-end-line="113" selection-end-column="54" />
531 |         </state>
532 |       </provider>
533 |     </entry>
534 |     <entry file="file://$PROJECT_DIR$/dog.0.jpg">
535 |       <provider selected="true" editor-type-id="images" />
536 |     </entry>
537 |     <entry file="file://$PROJECT_DIR$/lib/README.md" />
538 |     <entry file="file://$PROJECT_DIR$/lib/p3d_model.py" />
539 |     <entry file="file://$PROJECT_DIR$/.gitmodules">
540 |       <provider selected="true" editor-type-id="text-editor" />
541 |     </entry>
542 |     <entry file="file://$PROJECT_DIR$/utils.py">
543 |       <provider selected="true" editor-type-id="text-editor">
544 |         <state relative-caret-position="300">
545 |           <caret line="20" column="14" selection-start-line="20" selection-start-column="14" selection-end-line="20" selection-end-column="14" />
546 |         </state>
547 |       </provider>
548 |     </entry>
549 |     <entry file="file://$PROJECT_DIR$/test.py">
550 |       <provider selected="true" editor-type-id="text-editor">
551 |         <state relative-caret-position="1530">
552 |           <caret line="120" column="38" selection-start-line="120" selection-start-column="23" selection-end-line="120" selection-end-column="38" />
553 |           <folding>
554 |             <element signature="e#0#37#0" expanded="true" />
555 |           </folding>
556 |         </state>
557 |       </provider>
558 |     </entry>
559 |     <entry file="file://$PROJECT_DIR$/.gitignore">
560 |       <provider selected="true" editor-type-id="text-editor">
561 |         <state relative-caret-position="120">
562 |           <caret line="8" column="5" selection-start-line="8" selection-start-column="5" selection-end-line="8" selection-end-column="5" />
563 |         </state>
564 |       </provider>
565 |     </entry>
566 |     <entry file="file://$PROJECT_DIR$/Dataset.py">
567 |       <provider selected="true" editor-type-id="text-editor">
568 |         <state relative-caret-position="495">
569 |           <caret line="38" column="19" selection-start-line="38" selection-start-column="19" selection-end-line="38" selection-end-column="19" />
570 |           <folding>
571 |             <element signature="e#0#21#0" expanded="true" />
572 |           </folding>
573 |         </state>
574 |       </provider>
575 |     </entry>
576 |     <entry file="file://$PROJECT_DIR$/models/p3d_model.py">
577 |       <provider selected="true" editor-type-id="text-editor">
578 |         <state relative-caret-position="807">
579 |           <caret line="388" column="47" lean-forward="true" selection-start-line="387" selection-start-column="4" selection-end-line="388" selection-end-column="47" />
580 |         </state>
581 |       </provider>
582 |     </entry>
583 |     <entry file="file://$PROJECT_DIR$/transforms.py">
584 |       <provider selected="true" editor-type-id="text-editor">
585 |         <state relative-caret-position="744">
586 |           <caret line="92" column="10" selection-start-line="92" selection-start-column="10" selection-end-line="92" selection-end-column="10" />
587 |           <folding>
588 |             <marker date="1536122069233" expanded="true" signature="435:696" ph="..." />
589 |           </folding>
590 |         </state>
591 |       </provider>
592 |     </entry>
593 |     <entry file="file://$PROJECT_DIR$/main.py">
594 |       <provider selected="true" editor-type-id="text-editor">
595 |         <state relative-caret-position="510">
596 |           <caret line="37" column="44" lean-forward="true" selection-start-line="37" selection-start-column="44" selection-end-line="37" selection-end-column="44" />
597 |         </state>
598 |       </provider>
599 |     </entry>
600 |     <entry file="file://$PROJECT_DIR$/visualize.py">
601 |       <provider selected="true" editor-type-id="text-editor">
602 |         <state relative-caret-position="195">
603 |           <caret line="13" column="41" selection-start-line="13" selection-start-column="37" selection-end-line="13" selection-end-column="41" />
604 |         </state>
605 |       </provider>
606 |     </entry>
607 |     <entry file="file://$PROJECT_DIR$/train.py">
608 |       <provider selected="true" editor-type-id="text-editor">
609 |         <state relative-caret-position="792">
610 |           <caret line="381" column="8" selection-start-line="381" selection-start-column="8" selection-end-line="381" selection-end-column="8" />
611 |           <folding>
612 |             <element signature="e#0#37#0" expanded="true" />
613 |             <marker date="1536296846868" expanded="true" signature="7009:7014" ph="..." />
614 |           </folding>
615 |         </state>
616 |       </provider>
617 |     </entry>
618 |   </component>
619 | </project>


--------------------------------------------------------------------------------
/Dataset.py:
--------------------------------------------------------------------------------
 1 | from PIL import Image
 2 | import os
 3 | import os.path
 4 | import glob
 5 | import numpy as np
 6 | from torch.utils.data import Dataset
 7 | import random
 8 | 
 9 | class MyDataset(Dataset):
10 |     def __init__(self, root_path, data_folder='train', name_list='ucfTrainTestlist', version=1, transform=None, num_frames=16, modality='RGB', random=False):
11 |         self.root_path = root_path
12 |         self.num_frames = num_frames
13 |         self.data_folder = data_folder
14 |         self.random = random
15 |         self.split_file = os.path.join(self.root_path, name_list,
16 |                                        str(data_folder) + 'list0' + str(version) + '.txt')
17 |         self.label_file = os.path.join(self.root_path, name_list, 'classInd.txt')
18 |         self.label_dict = self.get_labels()
19 | 
20 |         self.video_dict = self.get_video_list()
21 | 
22 |         self.version = version
23 |         self.transform = transform
24 | 
25 |     def get_video_list(self):
26 |         res = []
27 |         with open(self.split_file) as fin:
28 |             for line in list(fin):
29 |                 line = line.replace("\n", "")
30 |                 split = line.split(" ")
31 |                 # get number frames of each video
32 |                 video_path = split[0].split('.')[0]
33 |                 frames_path = os.path.join(self.root_path, self.data_folder, video_path)
34 |                 allfiles = glob.glob(frames_path + '/*.jpg')
35 |                 # remove video which has < 16 image frames
36 |                 if len(allfiles) >= self.num_frames:
37 |                     res.append(split[0])
38 |         return res
39 | 
40 |     # Get all labels from classInd.txt
41 |     def get_labels(self):
42 |         label_dict = {}
43 |         with open(self.label_file) as fin:
44 |             for row in list(fin):
45 |                 row = row.replace("\n", "").split(" ")
46 |                 # -1 because the index of array is start from 0
47 |                 label_dict[row[1]] = int(row[0]) - 1
48 |         return label_dict
49 | 
50 |     # Get all frame images of video
51 |     def get_all_images(self, dir, file_ext="jpg", sort_files=True):
52 |         allfiles = glob.glob(dir + '/*.' + file_ext)
53 |         if sort_files and len(allfiles) > 0:
54 |             allfiles = sorted(allfiles)
55 |         return allfiles
56 | 
57 |     def get_video_tensor(self, dir):
58 |         images = self.get_all_images(dir)
59 |         # print(dir)
60 |         # print(len(images))
61 |         seed = np.random.random_integers(0, len(images) - self.num_frames)  # random sampling
62 |         clip = list()
63 |         if self.random:
64 |             orders = list(range(len(images)))
65 |             random_picked = random.sample(orders, self.num_frames)
66 |             for i in range(self.num_frames):
67 |                 idx = random_picked[i]
68 |                 img = Image.open(images[idx])
69 |                 clip.append(img)
70 |         else:
71 |             for i in range(self.num_frames):
72 |                 img = Image.open(images[i + seed])
73 |                 clip.append(img)
74 |         clip = self.transform(clip)
75 |         return clip
76 | 
77 |     # stuff
78 |     def __getitem__(self, index):
79 |         video = self.video_dict[index]
80 |         video_path = video.split('.')[0]
81 | 
82 |         frames_path = os.path.join(self.root_path, self.data_folder, video_path)
83 |         clip = self.get_video_tensor(frames_path)
84 |         # get label name from video path
85 |         label_name = video_path.split('/')[0]
86 |         label_index = self.label_dict[label_name];
87 |         return (clip, label_index)
88 | 
89 |     def __len__(self):
90 |         return len(self.video_dict)
91 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | 
 2 | # Pseudo-3D Residual Networks
 3 | 
 4 | This repo implements the network structure of P3D[1] with PyTorch, pre-trained model weights are converted from caffemodel, which is supported from the [author's repo](https://github.com/ZhaofanQiu/pseudo-3d-residual-networks)
 5 | 
 6 | 
 7 | ### Requirements:
 8 | 
 9 | - pytorch
10 | - numpy
11 | - ffmpeg (for extract image frames from videos)
12 | 
13 | ### Pretrained weights
14 | 
15 | 1, P3D-199 trained on Kinetics dataset:
16 | 
17 |  [Google Drive url](https://drive.google.com/drive/folders/1u_l-yvhS0shpW6e0tCiqPE7Bd1qQZKdD)
18 |  
19 | 2, P3D-199 trianed on Kinetics Optical Flow (TVL1):
20 | 
21 |  [Google Drive url](https://drive.google.com/drive/folders/1u_l-yvhS0shpW6e0tCiqPE7Bd1qQZKdD)
22 | 
23 |  
24 | ### Prepare Dataset UCF101
25 | First, download the dataset from UCF into the data folder and then extract it.
26 | ```
27 | cd data && wget http://crcv.ucf.edu/data/UCF101/UCF101.rar
28 | unrar e UCF101.rar
29 | ```
30 | 
31 | Next, make 3 folders train, test and validation:
32 | ```
33 | mkdir train test validation
34 | ```
35 | Finally, run scripts to extract image frames from videos;
36 | ```
37 | python move.py
38 | python makeVideoFolder.py
39 | python extract.py
40 | ```
41 | 
42 | ### Run Code
43 | 1, For Training from scratch
44 | ```
45 | python main.py /path/data/
46 | ```
47 | 2, For Fine-tuning
48 | ```
49 | python main.py /path/data/ --pretrained
50 | ```
51 | 3, For Evaluate model
52 | ```
53 | python main.py /path/data/ --resume=checkpoint.pth.tar --evaluate
54 | ```
55 | 4, For testing model
56 | ```
57 | python main.py /path/data/ --test
58 | ```
59 | 
60 | ### Experiment Result From Us
61 | Dataset | Accuracy
62 | ---|---|
63 | UCF-101 | 81.6%
64 | MERL Shopping | 82.6%
65 | 
66 | Reference:
67 | 
68 |  [1][Learning Spatio-Temporal Representation with Pseudo-3D Residual,ICCV2017](http://openaccess.thecvf.com/content_iccv_2017/html/Qiu_Learning_Spatio-Temporal_Representation_ICCV_2017_paper.html)
69 | 


--------------------------------------------------------------------------------
/data/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/naviocean/pseudo-3d-pytorch/50297d11248630792709782f467982e80c281384/data/.DS_Store


--------------------------------------------------------------------------------
/data/breakfastTrainTestList/classInd.txt:
--------------------------------------------------------------------------------
 1 | 1 AddSaltnpepper
 2 | 2 AddTeabag
 3 | 3 ButterPan
 4 | 4 CrackEgg
 5 | 5 CutBun
 6 | 6 CutFruit
 7 | 7 CutOrange
 8 | 8 FryEgg
 9 | 9 FryPancake
10 | 10 PeelFruit
11 | 11 PourCereals
12 | 12 PourCoffee
13 | 13 PourDough2pan
14 | 14 PourJuice
15 | 15 PourMilk
16 | 16 PourOil
17 | 17 PourWater
18 | 18 PutEgg2plate
19 | 19 PutFruit2bowl
20 | 20 PutToppingontop
21 | 21 SmearButter
22 | 22 SpoonPowder
23 | 23 SqueezeOrange
24 | 24 StirDough
25 | 25 StirMilk
26 | 26 TakeCup
27 | 27 TakePlate
28 | 


--------------------------------------------------------------------------------
/data/breakfastTrainTestList/fixclass.py:
--------------------------------------------------------------------------------
 1 | from random import shuffle
 2 | 
 3 | 
 4 | class FixClass(object):
 5 |     class_list = []
 6 |     train_file = './trainlist01.txt'
 7 |     class_file = './classInd.txt'
 8 | 
 9 |     def __init__(self):
10 |         super(FixClass, self).__init__()
11 |         # self.readFile()
12 |         # self.writeClassFile()
13 |         self.shuffeFile()
14 | 
15 |     def readFile(self):
16 |         with open(self.train_file) as fin:
17 |             for row in list(fin):
18 |                 class_name = row.strip().split('/')[0]
19 |                 if class_name not in self.class_list:
20 |                     self.class_list.append(class_name)
21 |         self.class_list.sort()
22 |         print(len(self.class_list))
23 | 
24 |     def writeClassFile(self):
25 |         # earse content file
26 |         with open(self.class_file, 'w'): pass
27 |         # write content file
28 |         with open(self.class_file, 'a') as f:
29 |             for i, key in enumerate(self.class_list):
30 |                 line = str(i + 1) + ' ' + key + '\n';
31 |                 f.write(line)
32 |         f.close
33 |         print('write class file done')
34 | 
35 |     def shuffeFile(self):
36 |         files = ['./trainlist01.txt','testlist01.txt','validationlist01.txt']
37 | 
38 |         for i in range(len(files)):
39 |             file = files[i]
40 |             with open(file) as fin:
41 |                 rows = [row.strip() for row in list(fin)]
42 | 
43 |             shuffle(rows)
44 |             # earse content file
45 |             with open(file, 'w'): pass
46 |             # write content file
47 |             with open(file, 'a') as f:
48 |                 for row in rows:
49 |                     line = row + '\n';
50 |                     f.write(line)
51 |             f.close
52 |             print('write file done')
53 | 
54 | 
55 | FixClass()
56 | 


--------------------------------------------------------------------------------
/data/breakfastTrainTestList/testlist01.txt:
--------------------------------------------------------------------------------
  1 | PourMilk/P32_pancake_996_PourMilk.avi
  2 | CutOrange/P32_juice_86_CutOrange.avi
  3 | CutFruit/P35_salad_642_CutFruit.avi
  4 | PutFruit2bowl/P25_salad_2496_PutFruit2bowl.avi
  5 | SqueezeOrange/P35_juice_156_SqueezeOrange.avi
  6 | PourMilk/P03_cereal_428_PourMilk.avi
  7 | CutFruit/P35_salad_187_CutFruit.avi
  8 | PutToppingontop/P27_sandwich_1003_PutToppingontop.avi
  9 | PutFruit2bowl/P25_salad_1576_PutFruit2bowl.avi
 10 | PutToppingontop/P28_sandwich_1899_PutToppingontop.avi
 11 | TakePlate/P40_salad_26_TakePlate.avi
 12 | FryEgg/P29_friedegg_279_FryEgg.avi
 13 | CutFruit/P31_salad_1029_CutFruit.avi
 14 | StirMilk/P33_milk_453_StirMilk.avi
 15 | PutFruit2bowl/P27_salad_1936_PutFruit2bowl.avi
 16 | PourJuice/P36_juice_1216_PourJuice.avi
 17 | PeelFruit/P31_salad_1184_PeelFruit.avi
 18 | CutBun/P34_sandwich_41_CutBun.avi
 19 | PeelFruit/P27_salad_61_PeelFruit.avi
 20 | CutFruit/P32_salad_1863_CutFruit.avi
 21 | PutEgg2plate/P31_scrambledegg_4304_PutEgg2plate.avi
 22 | PutEgg2plate/P29_scrambledegg_2493_PutEgg2plate.avi
 23 | SmearButter/P28_sandwich_649_SmearButter.avi
 24 | PourMilk/P28_cereal_212_PourMilk.avi
 25 | SmearButter/P35_sandwich_348_SmearButter.avi
 26 | AddSaltnpepper/P35_scrambledegg_285_AddSaltnpepper.avi
 27 | PourCoffee/P25_coffee_1_PourCoffee.avi
 28 | StirMilk/P40_milk_602_StirMilk.avi
 29 | PourMilk/P34_milk_436_PourMilk.avi
 30 | CutOrange/P35_juice_86_CutOrange.avi
 31 | CutFruit/P33_salad_3725_CutFruit.avi
 32 | PutFruit2bowl/P35_salad_572_PutFruit2bowl.avi
 33 | TakePlate/P40_scrambledegg_1542_TakePlate.avi
 34 | SqueezeOrange/P29_juice_314_SqueezeOrange.avi
 35 | CutFruit/P27_salad_1496_CutFruit.avi
 36 | TakePlate/P27_scrambledegg_2879_TakePlate.avi
 37 | PeelFruit/P25_salad_146_PeelFruit.avi
 38 | PutEgg2plate/P36_friedegg_3766_PutEgg2plate.avi
 39 | FryPancake/P35_pancake_1238_FryPancake.avi
 40 | FryEgg/P25_scrambledegg_1568_FryEgg.avi
 41 | CutFruit/P27_salad_3866_CutFruit.avi
 42 | AddTeabag/P35_tea_50_AddTeabag.avi
 43 | PeelFruit/P40_salad_1026_PeelFruit.avi
 44 | CutFruit/P32_salad_2183_CutFruit.avi
 45 | SmearButter/P34_sandwich_371_SmearButter.avi
 46 | PourMilk/P35_milk_331_PourMilk.avi
 47 | ButterPan/P33_friedegg_35_ButterPan.avi
 48 | FryEgg/P36_scrambledegg_952_FryEgg.avi
 49 | CutOrange/P27_juice_71_CutOrange.avi
 50 | SpoonPowder/P29_milk_262_SpoonPowder.avi
 51 | PourWater/P27_tea_566_PourWater.avi
 52 | PourWater/P32_tea_260_PourWater.avi
 53 | PutFruit2bowl/P31_salad_1429_PutFruit2bowl.avi
 54 | PourMilk/P30_coffee_277_PourMilk.avi
 55 | PutToppingontop/P34_sandwich_581_PutToppingontop.avi
 56 | CrackEgg/P29_friedegg_10_CrackEgg.avi
 57 | SqueezeOrange/P40_juice_212_SqueezeOrange.avi
 58 | FryEgg/P33_friedegg_1086_FryEgg.avi
 59 | CutBun/P27_sandwich_69_CutBun.avi
 60 | AddTeabag/P31_tea_244_AddTeabag.avi
 61 | PeelFruit/P31_salad_454_PeelFruit.avi
 62 | PourOil/P37_pancake_1155_PourOil.avi
 63 | PourCoffee/P34_coffee_56_PourCoffee.avi
 64 | StirDough/P40_pancake_785_StirDough.avi
 65 | SpoonPowder/P35_milk_171_SpoonPowder.avi
 66 | PourMilk/P37_cereal_299_PourMilk.avi
 67 | PutEgg2plate/P33_friedegg_2501_PutEgg2plate.avi
 68 | FryEgg/P30_scrambledegg_1663_FryEgg.avi
 69 | CutBun/P10_sandwich_120_CutBun.avi
 70 | TakeCup/P03_milk_54_TakeCup.avi
 71 | PourCereals/P29_cereal_59_PourCereals.avi
 72 | CrackEgg/P35_friedegg_159_CrackEgg.avi
 73 | SpoonPowder/P31_milk_657_SpoonPowder.avi
 74 | PutEgg2plate/P35_scrambledegg_1875_PutEgg2plate.avi
 75 | TakePlate/P37_friedegg_1823_TakePlate.avi
 76 | PourMilk/P33_pancake_912_PourMilk.avi
 77 | PourWater/P29_tea_370_PourWater.avi
 78 | PutFruit2bowl/P31_salad_4917_PutFruit2bowl.avi
 79 | CrackEgg/P32_scrambledegg_2_CrackEgg.avi
 80 | PutFruit2bowl/P36_salad_5056_PutFruit2bowl.avi
 81 | CutBun/P35_sandwich_187_CutBun.avi
 82 | FryPancake/P36_pancake_2843_FryPancake.avi
 83 | PutFruit2bowl/P37_salad_1420_PutFruit2bowl.avi
 84 | PeelFruit/P25_salad_1116_PeelFruit.avi
 85 | PutToppingontop/P03_sandwich_2780_PutToppingontop.avi
 86 | PourDough2pan/P32_pancake_2131_PourDough2pan.avi
 87 | FryPancake/P29_pancake_3711_FryPancake.avi
 88 | CrackEgg/P29_scrambledegg_66_CrackEgg.avi
 89 | FryEgg/P33_scrambledegg_2655_FryEgg.avi
 90 | CrackEgg/P31_pancake_1_CrackEgg.avi
 91 | StirDough/P31_pancake_1356_StirDough.avi
 92 | AddSaltnpepper/P37_scrambledegg_438_AddSaltnpepper.avi
 93 | CrackEgg/P27_friedegg_270_CrackEgg.avi
 94 | AddSaltnpepper/P32_scrambledegg_449_AddSaltnpepper.avi
 95 | PourMilk/P32_milk_359_PourMilk.avi
 96 | PourMilk/P29_milk_42_PourMilk.avi
 97 | PourMilk/P03_coffee_378_PourMilk.avi
 98 | PourOil/P27_scrambledegg_1030_PourOil.avi
 99 | PourDough2pan/P37_pancake_1342_PourDough2pan.avi
100 | PourMilk/P33_milk_314_PourMilk.avi
101 | CutFruit/P29_salad_786_CutFruit.avi
102 | PutFruit2bowl/P25_salad_2156_PutFruit2bowl.avi
103 | ButterPan/P31_scrambledegg_435_ButterPan.avi
104 | SqueezeOrange/P31_juice_401_SqueezeOrange.avi
105 | CutFruit/P35_salad_1197_CutFruit.avi
106 | PourCoffee/P37_coffee_1_PourCoffee.avi
107 | TakeCup/P31_coffee_1_TakeCup.avi
108 | TakeCup/P03_tea_89_TakeCup.avi
109 | CrackEgg/P36_scrambledegg_310_CrackEgg.avi
110 | TakePlate/P32_friedegg_2046_TakePlate.avi
111 | CrackEgg/P34_scrambledegg_1_CrackEgg.avi
112 | CutFruit/P03_salad_853_CutFruit.avi
113 | AddSaltnpepper/P32_friedegg_2534_AddSaltnpepper.avi
114 | StirMilk/P35_milk_421_StirMilk.avi
115 | CrackEgg/P27_pancake_1_CrackEgg.avi
116 | FryEgg/P25_friedegg_970_FryEgg.avi
117 | PourMilk/P35_cereal_331_PourMilk.avi
118 | PourMilk/P34_coffee_195_PourMilk.avi
119 | PeelFruit/P03_salad_3310_PeelFruit.avi
120 | PourCereals/P37_cereal_39_PourCereals.avi
121 | FryEgg/P33_scrambledegg_525_FryEgg.avi
122 | PutFruit2bowl/P32_salad_2323_PutFruit2bowl.avi
123 | PutEgg2plate/P37_scrambledegg_1089_PutEgg2plate.avi
124 | PourCereals/P33_cereal_82_PourCereals.avi
125 | CutFruit/P28_salad_1615_CutFruit.avi
126 | PourJuice/P25_juice_459_PourJuice.avi
127 | PutFruit2bowl/P29_salad_2466_PutFruit2bowl.avi
128 | PeelFruit/P28_salad_1775_PeelFruit.avi
129 | AddSaltnpepper/P32_friedegg_1357_AddSaltnpepper.avi
130 | CutFruit/P25_salad_936_CutFruit.avi
131 | StirDough/P27_pancake_657_StirDough.avi
132 | PourOil/P36_friedegg_1_PourOil.avi
133 | SpoonPowder/P30_milk_21_SpoonPowder.avi
134 | ButterPan/P30_scrambledegg_40_ButterPan.avi
135 | CutFruit/P37_salad_2700_CutFruit.avi
136 | CrackEgg/P36_pancake_278_CrackEgg.avi
137 | AddSaltnpepper/P27_scrambledegg_184_AddSaltnpepper.avi
138 | SmearButter/P10_sandwich_455_SmearButter.avi
139 | PourDough2pan/P30_pancake_2601_PourDough2pan.avi
140 | FryEgg/P37_scrambledegg_681_FryEgg.avi
141 | AddTeabag/P36_tea_11_AddTeabag.avi
142 | PourOil/P34_friedegg_50_PourOil.avi
143 | StirMilk/P36_milk_569_StirMilk.avi
144 | CutFruit/P28_salad_10_CutFruit.avi
145 | PourJuice/P27_juice_1656_PourJuice.avi
146 | 


--------------------------------------------------------------------------------
/data/breakfastTrainTestList/trainlist01.txt:
--------------------------------------------------------------------------------
  1 | PutEgg2plate/P34_scrambledegg_3778_PutEgg2plate.avi
  2 | SqueezeOrange/P28_juice_530_SqueezeOrange.avi
  3 | PutToppingontop/P37_sandwich_986_PutToppingontop.avi
  4 | FryPancake/P40_pancake_1764_FryPancake.avi
  5 | PutFruit2bowl/P27_salad_4516_PutFruit2bowl.avi
  6 | CrackEgg/P25_friedegg_629_CrackEgg.avi
  7 | CutFruit/P37_salad_20_CutFruit.avi
  8 | CrackEgg/P25_scrambledegg_437_CrackEgg.avi
  9 | PeelFruit/P31_salad_4142_PeelFruit.avi
 10 | TakePlate/P35_scrambledegg_1505_TakePlate.avi
 11 | PourOil/P36_pancake_1957_PourOil.avi
 12 | PeelFruit/P31_salad_3607_PeelFruit.avi
 13 | TakeCup/P31_tea_34_TakeCup.avi
 14 | PutEgg2plate/P40_scrambledegg_1633_PutEgg2plate.avi
 15 | SpoonPowder/P32_milk_214_SpoonPowder.avi
 16 | CutFruit/P25_salad_2196_CutFruit.avi
 17 | PutEgg2plate/P30_scrambledegg_2488_PutEgg2plate.avi
 18 | CutFruit/P32_salad_2033_CutFruit.avi
 19 | StirDough/P27_pancake_1642_StirDough.avi
 20 | FryEgg/P31_friedegg_1868_FryEgg.avi
 21 | PourMilk/P32_milk_124_PourMilk.avi
 22 | CutFruit/P31_salad_4762_CutFruit.avi
 23 | TakePlate/P30_pancake_5210_TakePlate.avi
 24 | PourMilk/P40_pancake_631_PourMilk.avi
 25 | PutFruit2bowl/P03_salad_4110_PutFruit2bowl.avi
 26 | PourMilk/P37_pancake_459_PourMilk.avi
 27 | ButterPan/P32_pancake_1926_ButterPan.avi
 28 | SqueezeOrange/P33_juice_272_SqueezeOrange.avi
 29 | StirMilk/P34_milk_801_StirMilk.avi
 30 | TakeCup/P33_coffee_83_TakeCup.avi
 31 | AddSaltnpepper/P37_friedegg_335_AddSaltnpepper.avi
 32 | PourWater/P36_tea_81_PourWater.avi
 33 | AddSaltnpepper/P34_scrambledegg_184_AddSaltnpepper.avi
 34 | AddSaltnpepper/P27_friedegg_2053_AddSaltnpepper.avi
 35 | PourMilk/P25_milk_787_PourMilk.avi
 36 | CutBun/P37_sandwich_11_CutBun.avi
 37 | CutOrange/P37_juice_61_CutOrange.avi
 38 | PourCoffee/P03_coffee_199_PourCoffee.avi
 39 | CrackEgg/P40_pancake_1_CrackEgg.avi
 40 | SqueezeOrange/P25_juice_133_SqueezeOrange.avi
 41 | CrackEgg/P32_pancake_1_CrackEgg.avi
 42 | CutFruit/P03_salad_548_CutFruit.avi
 43 | SqueezeOrange/P37_juice_221_SqueezeOrange.avi
 44 | TakePlate/P29_scrambledegg_2363_TakePlate.avi
 45 | PourMilk/P27_milk_70_PourMilk.avi
 46 | ButterPan/P31_pancake_3216_ButterPan.avi
 47 | PourJuice/P35_juice_736_PourJuice.avi
 48 | FryEgg/P37_friedegg_626_FryEgg.avi
 49 | CutFruit/P25_salad_1446_CutFruit.avi
 50 | SqueezeOrange/P32_juice_241_SqueezeOrange.avi
 51 | PutFruit2bowl/P27_salad_3816_PutFruit2bowl.avi
 52 | SpoonPowder/P31_milk_1907_SpoonPowder.avi
 53 | PutFruit2bowl/P32_salad_3278_PutFruit2bowl.avi
 54 | PourOil/P36_scrambledegg_3_PourOil.avi
 55 | PutFruit2bowl/P27_salad_4046_PutFruit2bowl.avi
 56 | PutFruit2bowl/P28_salad_1265_PutFruit2bowl.avi
 57 | PourCereals/P32_cereal_123_PourCereals.avi
 58 | CutFruit/P27_salad_2501_CutFruit.avi
 59 | FryPancake/P32_pancake_2582_FryPancake.avi
 60 | PourOil/P03_friedegg_385_PourOil.avi
 61 | PourMilk/P29_pancake_1068_PourMilk.avi
 62 | PourCereals/P40_cereal_56_PourCereals.avi
 63 | StirMilk/P03_milk_754_StirMilk.avi
 64 | FryPancake/P30_pancake_3076_FryPancake.avi
 65 | CutFruit/P32_salad_2853_CutFruit.avi
 66 | AddTeabag/P32_tea_120_AddTeabag.avi
 67 | CutOrange/P33_juice_52_CutOrange.avi
 68 | CutFruit/P28_salad_520_CutFruit.avi
 69 | PourOil/P35_scrambledegg_668_PourOil.avi
 70 | CrackEgg/P33_friedegg_329_CrackEgg.avi
 71 | CutBun/P33_sandwich_201_CutBun.avi
 72 | StirDough/P30_pancake_1382_StirDough.avi
 73 | FryEgg/P30_scrambledegg_1988_FryEgg.avi
 74 | PutFruit2bowl/P31_salad_2334_PutFruit2bowl.avi
 75 | CutFruit/P30_salad_1582_CutFruit.avi
 76 | CutFruit/P33_salad_4230_CutFruit.avi
 77 | PourCoffee/P32_coffee_46_PourCoffee.avi
 78 | PutEgg2plate/P35_friedegg_2520_PutEgg2plate.avi
 79 | SpoonPowder/P36_milk_40_SpoonPowder.avi
 80 | AddSaltnpepper/P30_scrambledegg_1533_AddSaltnpepper.avi
 81 | PeelFruit/P29_salad_1616_PeelFruit.avi
 82 | StirMilk/P27_milk_580_StirMilk.avi
 83 | PutFruit2bowl/P40_salad_2071_PutFruit2bowl.avi
 84 | SpoonPowder/P03_milk_229_SpoonPowder.avi
 85 | StirDough/P35_pancake_543_StirDough.avi
 86 | CrackEgg/P32_friedegg_829_CrackEgg.avi
 87 | FryEgg/P40_friedegg_743_FryEgg.avi
 88 | FryEgg/P34_friedegg_647_FryEgg.avi
 89 | FryEgg/P36_friedegg_840_FryEgg.avi
 90 | CrackEgg/P30_scrambledegg_640_CrackEgg.avi
 91 | CutFruit/P33_salad_30_CutFruit.avi
 92 | CutFruit/P29_salad_2151_CutFruit.avi
 93 | PourMilk/P27_pancake_1504_PourMilk.avi
 94 | SmearButter/P03_sandwich_652_SmearButter.avi
 95 | PourJuice/P30_juice_867_PourJuice.avi
 96 | CrackEgg/P37_pancake_1_CrackEgg.avi
 97 | PutFruit2bowl/P37_salad_3240_PutFruit2bowl.avi
 98 | PeelFruit/P28_salad_140_PeelFruit.avi
 99 | PutToppingontop/P25_sandwich_1045_PutToppingontop.avi
100 | CutFruit/P31_salad_1749_CutFruit.avi
101 | PourJuice/P33_juice_812_PourJuice.avi
102 | PutFruit2bowl/P31_salad_3557_PutFruit2bowl.avi
103 | PourDough2pan/P35_pancake_1037_PourDough2pan.avi
104 | CrackEgg/P37_friedegg_126_CrackEgg.avi
105 | SpoonPowder/P25_milk_382_SpoonPowder.avi
106 | PeelFruit/P30_salad_1203_PeelFruit.avi
107 | CutOrange/P36_juice_21_CutOrange.avi
108 | PeelFruit/P33_salad_3960_PeelFruit.avi
109 | PutFruit2bowl/P03_salad_2633_PutFruit2bowl.avi
110 | PourCereals/P28_cereal_44_PourCereals.avi
111 | PeelFruit/P31_salad_4967_PeelFruit.avi
112 | PutToppingontop/P03_sandwich_1763_PutToppingontop.avi
113 | CrackEgg/P03_friedegg_599_CrackEgg.avi
114 | CutBun/P40_sandwich_86_CutBun.avi
115 | PutFruit2bowl/P31_salad_1154_PutFruit2bowl.avi
116 | CutFruit/P31_salad_5297_CutFruit.avi
117 | SmearButter/P29_sandwich_559_SmearButter.avi
118 | PourOil/P25_friedegg_8_PourOil.avi
119 | FryEgg/P37_scrambledegg_143_FryEgg.avi
120 | AddTeabag/P25_tea_156_AddTeabag.avi
121 | TakePlate/P32_juice_36_TakePlate.avi
122 | CutFruit/P32_salad_218_CutFruit.avi
123 | PourCereals/P35_cereal_169_PourCereals.avi
124 | PeelFruit/P31_salad_2377_PeelFruit.avi
125 | PutEgg2plate/P27_scrambledegg_2986_PutEgg2plate.avi
126 | ButterPan/P33_pancake_1487_ButterPan.avi
127 | SqueezeOrange/P36_juice_286_SqueezeOrange.avi
128 | PourWater/P30_tea_127_PourWater.avi
129 | SmearButter/P33_sandwich_396_SmearButter.avi
130 | PutFruit2bowl/P27_salad_2401_PutFruit2bowl.avi
131 | PourWater/P31_tea_424_PourWater.avi
132 | PutEgg2plate/P32_friedegg_3468_PutEgg2plate.avi
133 | CutFruit/P33_salad_3185_CutFruit.avi
134 | FryEgg/P27_friedegg_525_FryEgg.avi
135 | PourMilk/P34_cereal_297_PourMilk.avi
136 | TakePlate/P27_friedegg_2256_TakePlate.avi
137 | CutFruit/P03_salad_2960_CutFruit.avi
138 | SmearButter/P36_sandwich_405_SmearButter.avi
139 | PutToppingontop/P31_sandwich_2077_PutToppingontop.avi
140 | PourCoffee/P33_coffee_121_PourCoffee.avi
141 | TakePlate/P32_pancake_6005_TakePlate.avi
142 | CrackEgg/P31_friedegg_475_CrackEgg.avi
143 | PutEgg2plate/P33_scrambledegg_2969_PutEgg2plate.avi
144 | FryEgg/P30_scrambledegg_1316_FryEgg.avi
145 | CutBun/P36_sandwich_15_CutBun.avi
146 | CutFruit/P31_salad_3937_CutFruit.avi
147 | PutEgg2plate/P03_friedegg_4007_PutEgg2plate.avi
148 | SpoonPowder/P33_milk_65_SpoonPowder.avi
149 | PeelFruit/P35_salad_857_PeelFruit.avi
150 | CutFruit/P40_salad_1636_CutFruit.avi
151 | PutEgg2plate/P36_scrambledegg_1838_PutEgg2plate.avi
152 | CutFruit/P03_salad_3740_CutFruit.avi
153 | CutFruit/P32_salad_2368_CutFruit.avi
154 | PourJuice/P30_juice_522_PourJuice.avi
155 | PourMilk/P30_pancake_1211_PourMilk.avi
156 | FryPancake/P30_pancake_5634_FryPancake.avi
157 | ButterPan/P40_friedegg_22_ButterPan.avi
158 | PourMilk/P33_coffee_381_PourMilk.avi
159 | PeelFruit/P29_salad_1036_PeelFruit.avi
160 | SqueezeOrange/P30_juice_592_SqueezeOrange.avi
161 | PourMilk/P31_pancake_1039_PourMilk.avi
162 | FryPancake/P27_pancake_2784_FryPancake.avi
163 | CutFruit/P31_salad_2104_CutFruit.avi
164 | PutFruit2bowl/P35_salad_827_PutFruit2bowl.avi
165 | StirDough/P32_pancake_1119_StirDough.avi
166 | PutToppingontop/P40_sandwich_1006_PutToppingontop.avi
167 | PutEgg2plate/P34_friedegg_2323_PutEgg2plate.avi
168 | AddTeabag/P27_tea_331_AddTeabag.avi
169 | PutEgg2plate/P27_friedegg_2656_PutEgg2plate.avi
170 | TakeCup/P40_milk_7_TakeCup.avi
171 | PourCoffee/P31_coffee_161_PourCoffee.avi
172 | SqueezeOrange/P27_juice_226_SqueezeOrange.avi
173 | PourJuice/P29_juice_1173_PourJuice.avi
174 | PutFruit2bowl/P32_salad_2778_PutFruit2bowl.avi
175 | FryEgg/P03_friedegg_1382_FryEgg.avi
176 | PourCereals/P25_cereal_304_PourCereals.avi
177 | ButterPan/P32_pancake_758_ButterPan.avi
178 | FryPancake/P37_pancake_1608_FryPancake.avi
179 | StirMilk/P29_milk_451_StirMilk.avi
180 | SqueezeOrange/P30_juice_145_SqueezeOrange.avi
181 | PourCereals/P27_cereal_49_PourCereals.avi
182 | SpoonPowder/P34_milk_221_SpoonPowder.avi
183 | FryEgg/P35_scrambledegg_1589_FryEgg.avi
184 | PourCereals/P36_cereal_32_PourCereals.avi
185 | PourMilk/P37_coffee_714_PourMilk.avi
186 | PutEgg2plate/P25_friedegg_1742_PutEgg2plate.avi
187 | CutBun/P31_sandwich_157_CutBun.avi
188 | ButterPan/P32_friedegg_291_ButterPan.avi
189 | PourOil/P35_friedegg_49_PourOil.avi
190 | PeelFruit/P27_salad_3021_PeelFruit.avi
191 | CutFruit/P36_salad_4011_CutFruit.avi
192 | PourMilk/P35_coffee_203_PourMilk.avi
193 | CrackEgg/P33_scrambledegg_163_CrackEgg.avi
194 | TakePlate/P29_pancake_6866_TakePlate.avi
195 | CutBun/P29_sandwich_280_CutBun.avi
196 | PourWater/P37_tea_21_PourWater.avi
197 | StirDough/P30_pancake_2366_StirDough.avi
198 | SmearButter/P31_sandwich_577_SmearButter.avi
199 | FryPancake/P31_pancake_4511_FryPancake.avi
200 | PourCereals/P31_cereal_95_PourCereals.avi
201 | PourCoffee/P30_coffee_76_PourCoffee.avi
202 | CutOrange/P30_juice_20_CutOrange.avi
203 | AddTeabag/P33_tea_260_AddTeabag.avi
204 | PeelFruit/P37_salad_1600_PeelFruit.avi
205 | FryEgg/P34_scrambledegg_1486_FryEgg.avi
206 | AddSaltnpepper/P25_friedegg_804_AddSaltnpepper.avi
207 | TakePlate/P27_pancake_5820_TakePlate.avi
208 | TakeCup/P03_coffee_38_TakeCup.avi
209 | PourMilk/P32_cereal_278_PourMilk.avi
210 | SmearButter/P40_sandwich_376_SmearButter.avi
211 | PourMilk/P29_cereal_241_PourMilk.avi
212 | PutFruit2bowl/P32_salad_2153_PutFruit2bowl.avi
213 | CutBun/P28_sandwich_384_CutBun.avi
214 | TakePlate/P31_pancake_8433_TakePlate.avi
215 | AddTeabag/P29_tea_50_AddTeabag.avi
216 | CrackEgg/P34_friedegg_228_CrackEgg.avi
217 | TakePlate/P30_scrambledegg_2329_TakePlate.avi
218 | PourMilk/P30_pancake_2285_PourMilk.avi
219 | PourWater/P34_tea_417_PourWater.avi
220 | CrackEgg/P30_pancake_1_CrackEgg.avi
221 | CutFruit/P33_salad_1240_CutFruit.avi
222 | PourDough2pan/P33_pancake_1879_PourDough2pan.avi
223 | PutToppingontop/P33_sandwich_856_PutToppingontop.avi
224 | AddSaltnpepper/P36_friedegg_418_AddSaltnpepper.avi
225 | SmearButter/P32_sandwich_221_SmearButter.avi
226 | PourMilk/P33_cereal_213_PourMilk.avi
227 | CrackEgg/P36_friedegg_94_CrackEgg.avi
228 | SmearButter/P27_sandwich_288_SmearButter.avi
229 | PutToppingontop/P32_sandwich_796_PutToppingontop.avi
230 | PutEgg2plate/P40_friedegg_2104_PutEgg2plate.avi
231 | CutFruit/P36_salad_181_CutFruit.avi
232 | PourDough2pan/P31_pancake_3604_PourDough2pan.avi
233 | PeelFruit/P33_salad_3405_PeelFruit.avi
234 | StirMilk/P32_milk_324_StirMilk.avi
235 | PourMilk/P40_milk_72_PourMilk.avi
236 | PourWater/P03_tea_919_PourWater.avi
237 | StirDough/P29_pancake_1229_StirDough.avi
238 | PutFruit2bowl/P35_salad_1362_PutFruit2bowl.avi
239 | AddSaltnpepper/P30_scrambledegg_1832_AddSaltnpepper.avi
240 | StirMilk/P37_milk_551_StirMilk.avi
241 | PourOil/P27_friedegg_24_PourOil.avi
242 | PourMilk/P31_milk_972_PourMilk.avi
243 | PourJuice/P34_juice_630_PourJuice.avi
244 | FryEgg/P31_scrambledegg_1655_FryEgg.avi
245 | CutFruit/P27_salad_3681_CutFruit.avi
246 | AddSaltnpepper/P33_scrambledegg_2220_AddSaltnpepper.avi
247 | CrackEgg/P27_scrambledegg_11_CrackEgg.avi
248 | PourMilk/P31_coffee_441_PourMilk.avi
249 | PourMilk/P40_cereal_308_PourMilk.avi
250 | CrackEgg/P37_scrambledegg_1_CrackEgg.avi
251 | CutOrange/P28_juice_140_CutOrange.avi
252 | PutFruit2bowl/P32_salad_1813_PutFruit2bowl.avi
253 | FryEgg/P32_scrambledegg_1491_FryEgg.avi
254 | PourOil/P34_scrambledegg_1082_PourOil.avi
255 | PourMilk/P35_pancake_413_PourMilk.avi
256 | CrackEgg/P31_scrambledegg_1450_CrackEgg.avi
257 | PourDough2pan/P40_pancake_1604_PourDough2pan.avi
258 | PutToppingontop/P10_sandwich_875_PutToppingontop.avi
259 | CutFruit/P29_salad_1206_CutFruit.avi
260 | PutFruit2bowl/P40_salad_931_PutFruit2bowl.avi
261 | PourMilk/P36_pancake_547_PourMilk.avi
262 | AddTeabag/P34_tea_176_AddTeabag.avi
263 | PeelFruit/P27_salad_1416_PeelFruit.avi
264 | TakeCup/P30_coffee_1_TakeCup.avi
265 | SmearButter/P37_sandwich_161_SmearButter.avi
266 | ButterPan/P29_pancake_2464_ButterPan.avi
267 | FryEgg/P31_scrambledegg_2409_FryEgg.avi
268 | PutFruit2bowl/P29_salad_1466_PutFruit2bowl.avi
269 | SmearButter/P25_sandwich_265_SmearButter.avi
270 | PutEgg2plate/P29_friedegg_3435_PutEgg2plate.avi
271 | PourMilk/P36_cereal_354_PourMilk.avi
272 | PeelFruit/P33_salad_180_PeelFruit.avi
273 | PourMilk/P36_milk_209_PourMilk.avi
274 | PeelFruit/P33_salad_2020_PeelFruit.avi
275 | CutFruit/P25_salad_1626_CutFruit.avi
276 | StirMilk/P31_milk_1292_StirMilk.avi
277 | TakeCup/P33_tea_30_TakeCup.avi
278 | PeelFruit/P31_salad_1464_PeelFruit.avi
279 | PourJuice/P32_juice_1151_PourJuice.avi
280 | PeelFruit/P31_salad_1884_PeelFruit.avi
281 | TakePlate/P40_sandwich_291_TakePlate.avi
282 | CutFruit/P25_salad_71_CutFruit.avi
283 | PourMilk/P30_milk_211_PourMilk.avi
284 | PourJuice/P37_juice_1390_PourJuice.avi
285 | SpoonPowder/P40_milk_302_SpoonPowder.avi
286 | PourMilk/P27_coffee_555_PourMilk.avi
287 | CutOrange/P31_juice_21_CutOrange.avi
288 | AddSaltnpepper/P31_scrambledegg_2319_AddSaltnpepper.avi
289 | CrackEgg/P29_pancake_1_CrackEgg.avi
290 | CutBun/P25_sandwich_65_CutBun.avi
291 | CutOrange/P40_juice_46_CutOrange.avi
292 | FryEgg/P29_scrambledegg_734_FryEgg.avi
293 | CrackEgg/P35_pancake_1_CrackEgg.avi
294 | PutToppingontop/P36_sandwich_1015_PutToppingontop.avi
295 | AddSaltnpepper/P35_friedegg_457_AddSaltnpepper.avi
296 | PutToppingontop/P10_sandwich_1550_PutToppingontop.avi
297 | StirDough/P33_pancake_1052_StirDough.avi
298 | CrackEgg/P33_pancake_348_CrackEgg.avi
299 | StirDough/P30_pancake_1938_StirDough.avi
300 | PourCoffee/P35_coffee_68_PourCoffee.avi
301 | AddTeabag/P30_tea_22_AddTeabag.avi
302 | FryEgg/P40_friedegg_1642_FryEgg.avi
303 | PutFruit2bowl/P31_salad_1854_PutFruit2bowl.avi
304 | PutFruit2bowl/P32_salad_1993_PutFruit2bowl.avi
305 | AddTeabag/P03_tea_239_AddTeabag.avi
306 | ButterPan/P32_scrambledegg_738_ButterPan.avi
307 | StirMilk/P31_milk_2107_StirMilk.avi
308 | PourMilk/P03_milk_514_PourMilk.avi
309 | PourJuice/P28_juice_1910_PourJuice.avi
310 | PourDough2pan/P36_pancake_2242_PourDough2pan.avi
311 | PourCereals/P03_cereal_150_PourCereals.avi
312 | PutEgg2plate/P37_friedegg_1868_PutEgg2plate.avi
313 | PourWater/P35_tea_285_PourWater.avi
314 | AddSaltnpepper/P34_friedegg_360_AddSaltnpepper.avi
315 | CutBun/P32_sandwich_21_CutBun.avi
316 | PourMilk/P25_cereal_484_PourMilk.avi
317 | StirDough/P37_pancake_575_StirDough.avi
318 | AddSaltnpepper/P36_scrambledegg_655_AddSaltnpepper.avi
319 | PutToppingontop/P35_sandwich_566_PutToppingontop.avi
320 | PourMilk/P27_cereal_449_PourMilk.avi
321 | CutFruit/P27_salad_2066_CutFruit.avi
322 | ButterPan/P31_friedegg_66_ButterPan.avi
323 | CutOrange/P29_juice_83_CutOrange.avi
324 | PourWater/P25_tea_886_PourWater.avi
325 | PourOil/P37_friedegg_25_PourOil.avi
326 | PeelFruit/P36_salad_1741_PeelFruit.avi
327 | TakeCup/P25_milk_232_TakeCup.avi
328 | PutFruit2bowl/P31_salad_5407_PutFruit2bowl.avi
329 | PutFruit2bowl/P28_salad_3945_PutFruit2bowl.avi
330 | FryEgg/P27_scrambledegg_1922_FryEgg.avi
331 | PourDough2pan/P27_pancake_2178_PourDough2pan.avi
332 | CutFruit/P40_salad_81_CutFruit.avi
333 | SpoonPowder/P27_milk_251_SpoonPowder.avi
334 | StirMilk/P30_milk_351_StirMilk.avi
335 | PourCoffee/P27_coffee_115_PourCoffee.avi
336 | CutFruit/P27_salad_1341_CutFruit.avi
337 | 


--------------------------------------------------------------------------------
/data/breakfastTrainTestList/validationlist01.txt:
--------------------------------------------------------------------------------
 1 | PourMilk/P31_cereal_347_PourMilk.avi
 2 | AddTeabag/P37_tea_201_AddTeabag.avi
 3 | AddSaltnpepper/P40_scrambledegg_399_AddSaltnpepper.avi
 4 | PourDough2pan/P29_pancake_3379_PourDough2pan.avi
 5 | PourCoffee/P29_coffee_1_PourCoffee.avi
 6 | StirMilk/P25_milk_912_StirMilk.avi
 7 | CrackEgg/P40_friedegg_504_CrackEgg.avi
 8 | PeelFruit/P29_salad_31_PeelFruit.avi
 9 | PutFruit2bowl/P31_salad_4102_PutFruit2bowl.avi
10 | CutOrange/P34_juice_75_CutOrange.avi
11 | SqueezeOrange/P34_juice_187_SqueezeOrange.avi
12 | CutFruit/P25_salad_746_CutFruit.avi
13 | CutFruit/P27_salad_4096_CutFruit.avi
14 | TakePlate/P31_scrambledegg_4090_TakePlate.avi
15 | CutFruit/P31_salad_1289_CutFruit.avi
16 | PourMilk/P37_cereal_357_PourMilk.avi
17 | CrackEgg/P35_scrambledegg_1_CrackEgg.avi
18 | PutFruit2bowl/P25_salad_886_PutFruit2bowl.avi
19 | PeelFruit/P30_salad_258_PeelFruit.avi
20 | FryEgg/P35_scrambledegg_1029_FryEgg.avi
21 | CrackEgg/P40_scrambledegg_1_CrackEgg.avi
22 | CutBun/P03_sandwich_260_CutBun.avi
23 | PourMilk/P29_coffee_316_PourMilk.avi
24 | AddSaltnpepper/P33_friedegg_768_AddSaltnpepper.avi
25 | CutFruit/P28_salad_3135_CutFruit.avi
26 | FryPancake/P33_pancake_2102_FryPancake.avi
27 | PourMilk/P37_milk_306_PourMilk.avi
28 | SmearButter/P10_sandwich_1145_SmearButter.avi
29 | ButterPan/P30_pancake_1640_ButterPan.avi
30 | PutEgg2plate/P31_friedegg_6459_PutEgg2plate.avi
31 | FryEgg/P35_friedegg_686_FryEgg.avi
32 | CutFruit/P30_salad_596_CutFruit.avi
33 | FryEgg/P40_scrambledegg_932_FryEgg.avi
34 | PourWater/P33_tea_100_PourWater.avi
35 | PutEgg2plate/P32_scrambledegg_2624_PutEgg2plate.avi
36 | SpoonPowder/P37_milk_21_SpoonPowder.avi
37 | PutToppingontop/P29_sandwich_1618_PutToppingontop.avi
38 | TakeCup/P34_tea_81_TakeCup.avi
39 | PutFruit2bowl/P36_salad_1571_PutFruit2bowl.avi
40 | StirDough/P36_pancake_1043_StirDough.avi
41 | PourOil/P35_pancake_961_PourOil.avi
42 | PourCereals/P34_cereal_63_PourCereals.avi
43 | PourJuice/P31_juice_1406_PourJuice.avi
44 | 


--------------------------------------------------------------------------------
/data/extract.py:
--------------------------------------------------------------------------------
  1 | """
  2 | After moving all the files using the 1_ file, we run this one to extract
  3 | the images from the videos and also create a data file we can use
  4 | for training and testing later.
  5 | """
  6 | import csv
  7 | import glob
  8 | import os
  9 | import os.path
 10 | from subprocess import call
 11 | 
 12 | 
 13 | def extract_files():
 14 |     """After we have all of our videos split between train and test, and
 15 |     all nested within folders representing their classes, we need to
 16 |     make a data file that we can reference when training our RNN(s).
 17 |     This will let us keep track of image sequences and other parts
 18 |     of the training process.
 19 | 
 20 |     We'll first need to extract images from each of the videos. We'll
 21 |     need to record the following data in the file:
 22 | 
 23 |     [train|test], class, filename, nb frames
 24 | 
 25 |     Extracting can be done with ffmpeg:
 26 |     `ffmpeg -i video.mpg image-%04d.jpg`
 27 |     """
 28 |     data_file = []
 29 |     folders = ['./train/', './test/', './validation/']
 30 | 
 31 |     for folder in folders:
 32 |         class_folders = glob.glob(folder + '*')
 33 |         for class_folder in class_folders:
 34 |             label_folders = glob.glob(class_folder + '/*')
 35 |             for vid_class in label_folders:
 36 |                 class_files = glob.glob(vid_class + '/*.avi')
 37 | 
 38 |                 for video_path in class_files:
 39 |                     # Get the parts of the file.
 40 |                     video_parts = get_video_parts(video_path)
 41 | 
 42 |                     train_or_test, classname, videofoldername, filename_no_ext, filename = video_parts
 43 | 
 44 |                     # Only extract if we haven't done it yet. Otherwise, just get
 45 |                     # the info.
 46 |                     if not check_already_extracted(video_parts):
 47 |                         # Now extract it.
 48 |                         src = train_or_test + '/' + classname + '/' + videofoldername + '/' +\
 49 |                             filename
 50 |                         dest = train_or_test + '/' + classname + '/' + videofoldername + '/' +\
 51 |                             filename_no_ext + '-%04d.jpg'
 52 |                         call(["ffmpeg", "-i", src, dest])
 53 | 
 54 |                     # Now get how many frames it is.
 55 |                     nb_frames = get_nb_frames_for_video(video_parts)
 56 | 
 57 |                     data_file.append(
 58 |                         [train_or_test, classname, filename_no_ext, nb_frames])
 59 | 
 60 |                     print("Generated %d frames for %s" %
 61 |                           (nb_frames, filename_no_ext))
 62 | 
 63 |     with open('data_file.csv', 'w') as fout:
 64 |         writer = csv.writer(fout)
 65 |         writer.writerows(data_file)
 66 | 
 67 |     print("Extracted and wrote %d video files." % (len(data_file)))
 68 | 
 69 | 
 70 | def get_nb_frames_for_video(video_parts):
 71 |     """Given video parts of an (assumed) already extracted video, return
 72 |     the number of frames that were extracted."""
 73 |     train_or_test, classname, videofoldername, filename_no_ext, _ = video_parts
 74 |     generated_files = glob.glob(train_or_test + '/' + classname + '/' + videofoldername + '/' +
 75 |                                 filename_no_ext + '*.jpg')
 76 |     return len(generated_files)
 77 | 
 78 | 
 79 | def get_video_parts(video_path):
 80 |     """Given a full path to a video, return its parts."""
 81 |     parts = video_path.split('/')
 82 |     filename = parts[4]
 83 |     filename_no_ext = filename.split('.')[0]
 84 |     videofoldername = parts[3]
 85 |     classname = parts[2]
 86 |     train_or_test = parts[1]
 87 | 
 88 |     return train_or_test, classname, videofoldername, filename_no_ext, filename
 89 | 
 90 | 
 91 | def check_already_extracted(video_parts):
 92 |     """Check to see if we created the -0001 frame of this file."""
 93 |     train_or_test, classname, videofoldername, filename_no_ext, _ = video_parts
 94 |     return bool(os.path.exists(train_or_test + '/' + classname + '/' + videofoldername +
 95 |                                '/' + filename_no_ext + '-0001.jpg'))
 96 | 
 97 | 
 98 | def main():
 99 |     """
100 |     Extract images from videos and build a new file that we
101 |     can use as our data input file. It can have format:
102 | 
103 |     [train|test], class, filename, nb frames
104 |     """
105 |     extract_files()
106 | 
107 | if __name__ == '__main__':
108 |     main()
109 | 


--------------------------------------------------------------------------------
/data/makeVideoFolder.py:
--------------------------------------------------------------------------------
 1 | import glob
 2 | import os
 3 | import os.path
 4 | import sys
 5 | import shutil
 6 | 
 7 | 
 8 | class MakeVideoFolder(object):
 9 |     """docstring for MakeVideoFolder"""
10 |     folders = ['./train/', './test/', './validation/']
11 | 
12 |     def __init__(self):
13 |         super(MakeVideoFolder, self).__init__()
14 |         self.scanFolder()
15 | 
16 |     def scanFolder(self):
17 |         for folder in self.folders:
18 |             class_folders = glob.glob(folder + '*')
19 |             for vid_class in class_folders:
20 |                 class_files = glob.glob(vid_class + '/*.*')
21 |                 for video_path in class_files:
22 |                     classname, filename_no_ext, filename = self.get_video_parts(
23 |                         video_path)
24 |                     folder_name = filename_no_ext.split('-')[0]
25 |                     destFolder = vid_class + '/' + folder_name
26 |                     self.checkAndMakeFolder(destFolder)
27 | 
28 |                     shutil.move(video_path, destFolder + '/' + filename)
29 |                     print('move file %s' % (filename_no_ext))
30 | 
31 |     def checkAndMakeFolder(self, folderPath):
32 |         try:
33 |             os.stat(folderPath)
34 |         except:
35 |             os.mkdir(folderPath)
36 | 
37 |     def get_video_parts(self, video_path):
38 |         """Given a full path to a video, return its parts."""
39 |         parts = video_path.split('/')
40 |         filename = parts[3]
41 |         filename_no_ext = filename.split('.')[0]
42 |         classname = parts[2]
43 |         return classname, filename_no_ext, filename
44 | 
45 | 
46 | MakeVideoFolder()
47 | 


--------------------------------------------------------------------------------
/data/merlTrainTestList/classInd.txt:
--------------------------------------------------------------------------------
1 | 1 ReachToShelf
2 | 2 RetractFromShelf
3 | 3 HandInShelf
4 | 4 InspectProduct
5 | 5 InspectShelf
6 | 


--------------------------------------------------------------------------------
/data/merlTrainTestList/merl.py:
--------------------------------------------------------------------------------
  1 | import scipy.io
  2 | import random
  3 | import math
  4 | import sys
  5 | import glob
  6 | import os
  7 | import os.path
  8 | from subprocess import call
  9 | 
 10 | 
 11 | class MERL_Shopping(object):
 12 |     classFile = './merlTrainTestList/classInd.txt'
 13 |     testFile = './merlTrainTestList/testlist01.txt'
 14 |     valFile = './merlTrainTestList/validationlist01.txt'
 15 |     trainFile = './merlTrainTestList/trainlist01.txt'
 16 |     inputVids = './Videos_MERL_Shopping_Dataset'
 17 |     outputVids = './output'
 18 |     labelFolder = './Labels_MERL_Shopping_Dataset'
 19 |     categories = dict()
 20 |     labels = ['ReachToShelf', 'RetractFromShelf',
 21 |               'HandInShelf', 'InspectProduct', 'InspectShelf']
 22 |     framerate = 30
 23 | 
 24 |     def __init__(self):
 25 |         self.checkDataFolder()
 26 | 
 27 |         self.checkOutputFolder()
 28 |         self.scanVideoFiles()
 29 |         self.pickRandom()
 30 | 
 31 |     def checkOutputFolder(self):
 32 |         try:
 33 |             os.stat(self.outputVids)
 34 |         except:
 35 |             os.mkdir(self.outputVids)
 36 | 
 37 |     def checkDataFolder(self):
 38 |         self.checkAndMakeFolder(folderPath='./merlTrainTestList')
 39 | 
 40 |     def checkAndMakeFolder(self, folderPath):
 41 |         try:
 42 |             os.stat(folderPath)
 43 |         except:
 44 |             os.mkdir(folderPath)
 45 | 
 46 |     def writeClassFile(self):
 47 |         # earse content file
 48 |         with open(self.classFile, 'w'):
 49 |             pass
 50 |         # write content file
 51 |         with open(self.classFile, 'a') as f:
 52 |             for i, key in enumerate(self.categories.keys()):
 53 |                 line = str(i + 1) + ' ' + key + '\n'
 54 |                 f.write(line)
 55 |         f.close
 56 |         print('write class file done')
 57 | 
 58 |     def writeOutputFile(self, list, file):
 59 |         # earse content file
 60 |         with open(file, 'w'):
 61 |             pass
 62 |         # write content file
 63 |         with open(file, 'a') as f:
 64 |             for filepath in list:
 65 |                 filename = os.path.basename(filepath)
 66 |                 cat = filename.split('.')[0]
 67 |                 cat = cat.split('_')
 68 |                 cat = cat[len(cat) - 1]
 69 |                 line = cat + '/' + filename + '\n'
 70 |                 f.write(line)
 71 |         f.close
 72 |         print('write file done')
 73 | 
 74 |     def checkFileExisted(self, filepath):
 75 |         return bool(os.path.exists(filepath))
 76 | 
 77 |     def checkAlreadySplit(self):
 78 |         return
 79 | 
 80 |     def spiltVideoByLabel(self, labelFilePath, src):
 81 |         # read matlab file
 82 |         mat = scipy.io.loadmat(labelFilePath)
 83 |         filename = os.path.basename(src)
 84 |         for i in range(len(mat['tlabs'])):
 85 |             label = self.labels[i]
 86 |             frames = mat['tlabs'][i][0]
 87 |             for i in range(len(frames)):
 88 |                 # get start frame
 89 |                 start_frame = int(frames[i][0])
 90 |                 start_time = start_frame / self.framerate
 91 |                 # get end frame
 92 |                 end_frame = int(frames[i][1])
 93 |                 end_time = (end_frame - start_frame) / self.framerate
 94 | 
 95 |                 folder = self.outputVids + '/' + label
 96 |                 dest_name = filename.split(
 97 |                     '.')[0] + '_' + str(start_frame) + '_' + str(end_frame) + '_' + label + '.mp4'
 98 | 
 99 |                 dest = folder + '/' + dest_name
100 | 
101 |                 # check folder existed or not
102 |                 self.checkAndMakeFolder(folder)
103 |                 if not self.checkFileExisted(filepath=dest):
104 |                     call(["ffmpeg", '-i', src, '-ss',
105 |                           str(start_time), '-t', str(end_time), dest])
106 |                     print("Splited %s" % (src))
107 |                     print(start_frame, end_frame, start_time, end_time, dest)
108 | 
109 |                 if label in self.categories:
110 |                     self.categories[label].append(
111 |                         label + '/' + dest_name)
112 |                 else:
113 |                     self.categories[label] = [
114 |                         label + '/' + dest_name]
115 | 
116 |     # get all video files
117 |     # read coarse file and then split video to multiparts with label
118 |     def scanVideoFiles(self):
119 |         for root, dirs, files in os.walk(self.inputVids):
120 |             for file in files:
121 |                 if file.endswith(".mp4"):
122 |                     mp4FilePath = root + '/' + file
123 |                     filename = file.split('.')[0]
124 |                     filename = filename.replace('_crop', '_label')
125 |                     # generate coarse file path
126 |                     labelFilePath = self.labelFolder + '/' + filename + '.mat'
127 |                     # check coarse file existed or not?
128 |                     if self.checkFileExisted(filepath=labelFilePath):
129 |                         # read coarse file to get label
130 |                         self.spiltVideoByLabel(
131 |                             labelFilePath, src=mp4FilePath)
132 |                     else:
133 |                         print('not found %s %s' % (root, mp4FilePath))
134 | 
135 |     def diff(self, li1, li2):
136 |         li_dif = [i for i in li1 + li2 if i not in li1 or i not in li2]
137 |         return li_dif
138 | 
139 |     def getRandomList(self, dataList):
140 |         if dataList != []:
141 |             elem = random.choice(dataList)
142 |             dataList.remove(elem)
143 |         else:
144 |             elem = None
145 |             # print(dataList)
146 |         return dataList, elem
147 | 
148 |     def getNumRandomList(self, datalist, num):
149 |         newlist = []
150 |         for x in range(0, num):
151 |             datalist, item = self.getRandomList(datalist)
152 |             newlist.append(item)
153 |         return datalist, newlist
154 | 
155 |     def pickRandom(self):
156 |         val_list = []
157 |         test_list = []
158 |         train_list = []
159 |         # print("%s" % (self.categories))
160 |         cats = self.categories
161 |         for key in cats.keys():
162 |             total = len(cats[key])
163 |             if total >= 10:
164 |                 val_total = math.trunc(total * (1 / 10))
165 |                 if val_total == 0:
166 |                     val_total = 1
167 |                 test_total = math.trunc(total * (3 / 10))
168 |                 if test_total == 0:
169 |                     test_total = 1
170 |                 train_total = total - val_total - test_total
171 |                 cats[key], tmp_list = self.getNumRandomList(
172 |                     cats[key], val_total)
173 | 
174 |                 val_list = val_list + tmp_list
175 | 
176 |                 cats[key], tmp_list = self.getNumRandomList(
177 |                     cats[key], test_total)
178 |                 test_list = test_list + tmp_list
179 |                 train_list = train_list + cats[key]
180 |                 print(total, val_total, test_total, train_total)
181 |         # print(test_list)
182 |         if len(val_list) > 0:
183 |             self.writeOutputFile(val_list, self.valFile)
184 |         if len(test_list) > 0:
185 |             self.writeOutputFile(test_list, self.testFile)
186 |         if len(train_list) > 0:
187 |             self.writeOutputFile(train_list, self.trainFile)
188 | 
189 |         self.writeClassFile()
190 | 
191 | 
192 | MERL_Shopping()
193 | 


--------------------------------------------------------------------------------
/data/merlTrainTestList/validationlist01.txt:
--------------------------------------------------------------------------------
  1 | ReachToShelf/16_3_crop_1719_1746_ReachToShelf.mp4
  2 | ReachToShelf/10_3_crop_3966_3997_ReachToShelf.mp4
  3 | ReachToShelf/11_1_crop_2177_2203_ReachToShelf.mp4
  4 | ReachToShelf/7_2_crop_1967_1999_ReachToShelf.mp4
  5 | ReachToShelf/17_2_crop_2647_2669_ReachToShelf.mp4
  6 | ReachToShelf/17_2_crop_1577_1605_ReachToShelf.mp4
  7 | ReachToShelf/37_1_crop_3266_3301_ReachToShelf.mp4
  8 | ReachToShelf/40_1_crop_526_561_ReachToShelf.mp4
  9 | ReachToShelf/7_3_crop_1672_1707_ReachToShelf.mp4
 10 | ReachToShelf/1_1_crop_376_406_ReachToShelf.mp4
 11 | ReachToShelf/13_2_crop_4061_4144_ReachToShelf.mp4
 12 | ReachToShelf/9_1_crop_605_636_ReachToShelf.mp4
 13 | ReachToShelf/19_1_crop_2410_2430_ReachToShelf.mp4
 14 | ReachToShelf/4_2_crop_1036_1084_ReachToShelf.mp4
 15 | ReachToShelf/17_1_crop_2034_2055_ReachToShelf.mp4
 16 | ReachToShelf/16_1_crop_2479_2497_ReachToShelf.mp4
 17 | ReachToShelf/19_2_crop_2682_2710_ReachToShelf.mp4
 18 | ReachToShelf/27_2_crop_1876_1911_ReachToShelf.mp4
 19 | ReachToShelf/17_2_crop_2563_2575_ReachToShelf.mp4
 20 | ReachToShelf/36_1_crop_1546_1576_ReachToShelf.mp4
 21 | ReachToShelf/33_1_crop_3105_3140_ReachToShelf.mp4
 22 | ReachToShelf/13_3_crop_607_639_ReachToShelf.mp4
 23 | ReachToShelf/1_1_crop_156_183_ReachToShelf.mp4
 24 | ReachToShelf/32_3_crop_1019_1035_ReachToShelf.mp4
 25 | ReachToShelf/4_2_crop_3487_3513_ReachToShelf.mp4
 26 | ReachToShelf/16_3_crop_2616_2650_ReachToShelf.mp4
 27 | ReachToShelf/26_3_crop_947_973_ReachToShelf.mp4
 28 | ReachToShelf/8_1_crop_271_297_ReachToShelf.mp4
 29 | ReachToShelf/1_1_crop_3377_3414_ReachToShelf.mp4
 30 | ReachToShelf/2_3_crop_2343_2384_ReachToShelf.mp4
 31 | ReachToShelf/23_2_crop_4669_4694_ReachToShelf.mp4
 32 | ReachToShelf/15_2_crop_230_260_ReachToShelf.mp4
 33 | ReachToShelf/37_1_crop_200_243_ReachToShelf.mp4
 34 | ReachToShelf/10_3_crop_480_506_ReachToShelf.mp4
 35 | ReachToShelf/12_1_crop_193_224_ReachToShelf.mp4
 36 | ReachToShelf/8_1_crop_2733_2773_ReachToShelf.mp4
 37 | ReachToShelf/2_3_crop_1176_1205_ReachToShelf.mp4
 38 | ReachToShelf/21_3_crop_2551_2583_ReachToShelf.mp4
 39 | ReachToShelf/5_1_crop_2198_2212_ReachToShelf.mp4
 40 | ReachToShelf/38_1_crop_1322_1359_ReachToShelf.mp4
 41 | ReachToShelf/10_2_crop_1222_1255_ReachToShelf.mp4
 42 | ReachToShelf/19_3_crop_3524_3556_ReachToShelf.mp4
 43 | ReachToShelf/20_3_crop_349_387_ReachToShelf.mp4
 44 | ReachToShelf/16_2_crop_3682_3731_ReachToShelf.mp4
 45 | ReachToShelf/19_1_crop_180_238_ReachToShelf.mp4
 46 | ReachToShelf/29_3_crop_751_776_ReachToShelf.mp4
 47 | ReachToShelf/17_2_crop_3955_4004_ReachToShelf.mp4
 48 | ReachToShelf/20_3_crop_2700_2733_ReachToShelf.mp4
 49 | ReachToShelf/28_3_crop_750_782_ReachToShelf.mp4
 50 | ReachToShelf/11_2_crop_699_749_ReachToShelf.mp4
 51 | ReachToShelf/18_1_crop_1430_1465_ReachToShelf.mp4
 52 | ReachToShelf/21_3_crop_558_584_ReachToShelf.mp4
 53 | ReachToShelf/20_1_crop_1687_1731_ReachToShelf.mp4
 54 | ReachToShelf/7_3_crop_3726_3762_ReachToShelf.mp4
 55 | ReachToShelf/26_3_crop_1164_1184_ReachToShelf.mp4
 56 | ReachToShelf/17_1_crop_391_414_ReachToShelf.mp4
 57 | ReachToShelf/18_1_crop_2158_2194_ReachToShelf.mp4
 58 | ReachToShelf/22_2_crop_3555_3583_ReachToShelf.mp4
 59 | ReachToShelf/38_1_crop_3328_3359_ReachToShelf.mp4
 60 | ReachToShelf/2_1_crop_292_327_ReachToShelf.mp4
 61 | ReachToShelf/9_1_crop_538_558_ReachToShelf.mp4
 62 | ReachToShelf/18_1_crop_770_804_ReachToShelf.mp4
 63 | ReachToShelf/30_1_crop_3520_3538_ReachToShelf.mp4
 64 | ReachToShelf/24_1_crop_3068_3086_ReachToShelf.mp4
 65 | ReachToShelf/26_1_crop_2622_2687_ReachToShelf.mp4
 66 | ReachToShelf/2_2_crop_1293_1322_ReachToShelf.mp4
 67 | ReachToShelf/31_1_crop_2340_2389_ReachToShelf.mp4
 68 | ReachToShelf/21_1_crop_2660_2678_ReachToShelf.mp4
 69 | ReachToShelf/28_2_crop_1448_1472_ReachToShelf.mp4
 70 | ReachToShelf/11_2_crop_581_613_ReachToShelf.mp4
 71 | ReachToShelf/31_2_crop_3409_3429_ReachToShelf.mp4
 72 | ReachToShelf/36_1_crop_2140_2170_ReachToShelf.mp4
 73 | ReachToShelf/31_1_crop_2769_2797_ReachToShelf.mp4
 74 | ReachToShelf/19_3_crop_3231_3258_ReachToShelf.mp4
 75 | ReachToShelf/13_2_crop_2514_2530_ReachToShelf.mp4
 76 | ReachToShelf/29_3_crop_5169_5208_ReachToShelf.mp4
 77 | ReachToShelf/32_3_crop_624_646_ReachToShelf.mp4
 78 | ReachToShelf/12_3_crop_2545_2586_ReachToShelf.mp4
 79 | ReachToShelf/15_2_crop_2678_2704_ReachToShelf.mp4
 80 | ReachToShelf/19_1_crop_571_594_ReachToShelf.mp4
 81 | ReachToShelf/27_2_crop_1660_1690_ReachToShelf.mp4
 82 | ReachToShelf/15_3_crop_836_872_ReachToShelf.mp4
 83 | ReachToShelf/13_2_crop_2801_2852_ReachToShelf.mp4
 84 | ReachToShelf/10_2_crop_1030_1049_ReachToShelf.mp4
 85 | ReachToShelf/24_3_crop_1279_1313_ReachToShelf.mp4
 86 | ReachToShelf/11_3_crop_2446_2478_ReachToShelf.mp4
 87 | ReachToShelf/17_2_crop_2439_2455_ReachToShelf.mp4
 88 | ReachToShelf/41_1_crop_3997_4068_ReachToShelf.mp4
 89 | ReachToShelf/10_3_crop_2600_2638_ReachToShelf.mp4
 90 | ReachToShelf/17_1_crop_1175_1217_ReachToShelf.mp4
 91 | ReachToShelf/6_3_crop_2340_2378_ReachToShelf.mp4
 92 | ReachToShelf/3_3_crop_1553_1579_ReachToShelf.mp4
 93 | ReachToShelf/2_3_crop_2259_2282_ReachToShelf.mp4
 94 | ReachToShelf/12_2_crop_293_331_ReachToShelf.mp4
 95 | ReachToShelf/5_3_crop_1273_1307_ReachToShelf.mp4
 96 | ReachToShelf/6_1_crop_456_492_ReachToShelf.mp4
 97 | ReachToShelf/18_3_crop_1447_1484_ReachToShelf.mp4
 98 | ReachToShelf/20_3_crop_976_994_ReachToShelf.mp4
 99 | ReachToShelf/32_1_crop_3753_3806_ReachToShelf.mp4
100 | ReachToShelf/21_1_crop_1683_1700_ReachToShelf.mp4
101 | ReachToShelf/1_3_crop_3410_3440_ReachToShelf.mp4
102 | ReachToShelf/20_3_crop_552_609_ReachToShelf.mp4
103 | ReachToShelf/40_1_crop_3550_3589_ReachToShelf.mp4
104 | ReachToShelf/8_2_crop_3838_3876_ReachToShelf.mp4
105 | ReachToShelf/27_3_crop_3685_3733_ReachToShelf.mp4
106 | ReachToShelf/10_2_crop_762_797_ReachToShelf.mp4
107 | ReachToShelf/9_1_crop_913_935_ReachToShelf.mp4
108 | ReachToShelf/1_3_crop_829_853_ReachToShelf.mp4
109 | ReachToShelf/3_2_crop_986_1009_ReachToShelf.mp4
110 | ReachToShelf/40_1_crop_1510_1547_ReachToShelf.mp4
111 | ReachToShelf/24_3_crop_4292_4313_ReachToShelf.mp4
112 | ReachToShelf/14_2_crop_681_702_ReachToShelf.mp4
113 | ReachToShelf/19_1_crop_2920_2946_ReachToShelf.mp4
114 | ReachToShelf/19_3_crop_1218_1249_ReachToShelf.mp4
115 | ReachToShelf/17_2_crop_2215_2248_ReachToShelf.mp4
116 | ReachToShelf/28_3_crop_2657_2700_ReachToShelf.mp4
117 | ReachToShelf/4_2_crop_751_789_ReachToShelf.mp4
118 | ReachToShelf/2_3_crop_2525_2550_ReachToShelf.mp4
119 | ReachToShelf/10_2_crop_3142_3162_ReachToShelf.mp4
120 | ReachToShelf/12_1_crop_1377_1399_ReachToShelf.mp4
121 | ReachToShelf/33_1_crop_442_458_ReachToShelf.mp4
122 | ReachToShelf/30_3_crop_4033_4060_ReachToShelf.mp4
123 | ReachToShelf/27_3_crop_3459_3488_ReachToShelf.mp4
124 | ReachToShelf/21_2_crop_1687_1712_ReachToShelf.mp4
125 | ReachToShelf/6_1_crop_2441_2480_ReachToShelf.mp4
126 | ReachToShelf/41_2_crop_2532_2583_ReachToShelf.mp4
127 | ReachToShelf/7_1_crop_3128_3141_ReachToShelf.mp4
128 | ReachToShelf/23_2_crop_4630_4666_ReachToShelf.mp4
129 | ReachToShelf/32_1_crop_154_181_ReachToShelf.mp4
130 | ReachToShelf/30_3_crop_4138_4157_ReachToShelf.mp4
131 | ReachToShelf/10_3_crop_3616_3650_ReachToShelf.mp4
132 | ReachToShelf/18_1_crop_983_1036_ReachToShelf.mp4
133 | ReachToShelf/25_2_crop_3374_3402_ReachToShelf.mp4
134 | ReachToShelf/6_3_crop_815_850_ReachToShelf.mp4
135 | ReachToShelf/36_1_crop_1263_1296_ReachToShelf.mp4
136 | ReachToShelf/36_1_crop_897_934_ReachToShelf.mp4
137 | ReachToShelf/4_1_crop_294_327_ReachToShelf.mp4
138 | ReachToShelf/26_2_crop_1570_1611_ReachToShelf.mp4
139 | ReachToShelf/18_3_crop_2363_2396_ReachToShelf.mp4
140 | ReachToShelf/21_2_crop_2199_2224_ReachToShelf.mp4
141 | ReachToShelf/17_3_crop_1287_1324_ReachToShelf.mp4
142 | ReachToShelf/14_1_crop_3711_3744_ReachToShelf.mp4
143 | ReachToShelf/1_3_crop_1441_1462_ReachToShelf.mp4
144 | ReachToShelf/8_2_crop_1950_1979_ReachToShelf.mp4
145 | ReachToShelf/17_3_crop_4067_4101_ReachToShelf.mp4
146 | ReachToShelf/18_2_crop_524_545_ReachToShelf.mp4
147 | ReachToShelf/3_1_crop_1034_1061_ReachToShelf.mp4
148 | ReachToShelf/8_2_crop_2007_2034_ReachToShelf.mp4
149 | ReachToShelf/14_3_crop_394_419_ReachToShelf.mp4
150 | ReachToShelf/27_1_crop_2389_2419_ReachToShelf.mp4
151 | ReachToShelf/31_3_crop_1076_1105_ReachToShelf.mp4
152 | ReachToShelf/24_2_crop_3364_3393_ReachToShelf.mp4
153 | ReachToShelf/11_3_crop_3083_3107_ReachToShelf.mp4
154 | ReachToShelf/19_3_crop_1060_1086_ReachToShelf.mp4
155 | ReachToShelf/5_2_crop_1437_1471_ReachToShelf.mp4
156 | ReachToShelf/15_2_crop_1710_1740_ReachToShelf.mp4
157 | ReachToShelf/26_1_crop_4076_4102_ReachToShelf.mp4
158 | ReachToShelf/38_1_crop_3508_3563_ReachToShelf.mp4
159 | ReachToShelf/11_3_crop_2235_2252_ReachToShelf.mp4
160 | ReachToShelf/20_1_crop_923_941_ReachToShelf.mp4
161 | ReachToShelf/34_1_crop_3719_3766_ReachToShelf.mp4
162 | ReachToShelf/24_3_crop_4383_4404_ReachToShelf.mp4
163 | ReachToShelf/26_1_crop_2779_2812_ReachToShelf.mp4
164 | ReachToShelf/1_3_crop_2434_2469_ReachToShelf.mp4
165 | ReachToShelf/24_2_crop_3946_3981_ReachToShelf.mp4
166 | ReachToShelf/16_3_crop_468_483_ReachToShelf.mp4
167 | ReachToShelf/12_2_crop_1557_1582_ReachToShelf.mp4
168 | ReachToShelf/32_2_crop_3303_3327_ReachToShelf.mp4
169 | ReachToShelf/41_1_crop_2858_2880_ReachToShelf.mp4
170 | ReachToShelf/19_2_crop_3220_3234_ReachToShelf.mp4
171 | ReachToShelf/27_2_crop_2660_2678_ReachToShelf.mp4
172 | RetractFromShelf/15_3_crop_1007_1032_RetractFromShelf.mp4
173 | RetractFromShelf/22_3_crop_505_567_RetractFromShelf.mp4
174 | RetractFromShelf/2_2_crop_1323_1342_RetractFromShelf.mp4
175 | RetractFromShelf/25_1_crop_1262_1305_RetractFromShelf.mp4
176 | RetractFromShelf/2_1_crop_862_895_RetractFromShelf.mp4
177 | RetractFromShelf/13_2_crop_1367_1392_RetractFromShelf.mp4
178 | RetractFromShelf/20_3_crop_2115_2151_RetractFromShelf.mp4
179 | RetractFromShelf/15_2_crop_176_204_RetractFromShelf.mp4
180 | RetractFromShelf/18_2_crop_1033_1056_RetractFromShelf.mp4
181 | RetractFromShelf/18_2_crop_502_523_RetractFromShelf.mp4
182 | RetractFromShelf/18_2_crop_691_713_RetractFromShelf.mp4
183 | RetractFromShelf/37_1_crop_809_840_RetractFromShelf.mp4
184 | RetractFromShelf/18_1_crop_1652_1690_RetractFromShelf.mp4
185 | RetractFromShelf/28_3_crop_3356_3395_RetractFromShelf.mp4
186 | RetractFromShelf/39_1_crop_2377_2418_RetractFromShelf.mp4
187 | RetractFromShelf/6_3_crop_3731_3767_RetractFromShelf.mp4
188 | RetractFromShelf/31_1_crop_381_418_RetractFromShelf.mp4
189 | RetractFromShelf/13_3_crop_1777_1816_RetractFromShelf.mp4
190 | RetractFromShelf/25_2_crop_3843_3876_RetractFromShelf.mp4
191 | RetractFromShelf/22_2_crop_620_667_RetractFromShelf.mp4
192 | RetractFromShelf/30_3_crop_4061_4086_RetractFromShelf.mp4
193 | RetractFromShelf/18_2_crop_2504_2523_RetractFromShelf.mp4
194 | RetractFromShelf/2_2_crop_2907_2940_RetractFromShelf.mp4
195 | RetractFromShelf/26_1_crop_2514_2549_RetractFromShelf.mp4
196 | RetractFromShelf/11_1_crop_2506_2529_RetractFromShelf.mp4
197 | RetractFromShelf/7_1_crop_1741_1770_RetractFromShelf.mp4
198 | RetractFromShelf/14_3_crop_2606_2638_RetractFromShelf.mp4
199 | RetractFromShelf/36_1_crop_3029_3068_RetractFromShelf.mp4
200 | RetractFromShelf/12_1_crop_942_1004_RetractFromShelf.mp4
201 | RetractFromShelf/11_1_crop_2634_2671_RetractFromShelf.mp4
202 | RetractFromShelf/15_3_crop_1730_1787_RetractFromShelf.mp4
203 | RetractFromShelf/22_2_crop_2745_2807_RetractFromShelf.mp4
204 | RetractFromShelf/29_2_crop_1995_2026_RetractFromShelf.mp4
205 | RetractFromShelf/29_2_crop_3279_3313_RetractFromShelf.mp4
206 | RetractFromShelf/3_3_crop_879_932_RetractFromShelf.mp4
207 | RetractFromShelf/39_1_crop_343_420_RetractFromShelf.mp4
208 | RetractFromShelf/10_3_crop_2656_2694_RetractFromShelf.mp4
209 | RetractFromShelf/6_1_crop_587_639_RetractFromShelf.mp4
210 | RetractFromShelf/24_1_crop_2827_2859_RetractFromShelf.mp4
211 | RetractFromShelf/24_3_crop_4405_4474_RetractFromShelf.mp4
212 | RetractFromShelf/17_2_crop_3469_3485_RetractFromShelf.mp4
213 | RetractFromShelf/16_1_crop_2805_2822_RetractFromShelf.mp4
214 | RetractFromShelf/9_1_crop_2538_2601_RetractFromShelf.mp4
215 | RetractFromShelf/17_2_crop_1605_1637_RetractFromShelf.mp4
216 | RetractFromShelf/15_2_crop_2195_2213_RetractFromShelf.mp4
217 | RetractFromShelf/26_2_crop_3603_3633_RetractFromShelf.mp4
218 | RetractFromShelf/41_1_crop_1212_1250_RetractFromShelf.mp4
219 | RetractFromShelf/13_1_crop_3432_3469_RetractFromShelf.mp4
220 | RetractFromShelf/32_3_crop_2569_2616_RetractFromShelf.mp4
221 | RetractFromShelf/38_1_crop_1615_1664_RetractFromShelf.mp4
222 | RetractFromShelf/28_2_crop_3909_3953_RetractFromShelf.mp4
223 | RetractFromShelf/23_1_crop_1739_1798_RetractFromShelf.mp4
224 | RetractFromShelf/15_2_crop_996_1023_RetractFromShelf.mp4
225 | RetractFromShelf/28_1_crop_1691_1746_RetractFromShelf.mp4
226 | RetractFromShelf/6_3_crop_2379_2472_RetractFromShelf.mp4
227 | RetractFromShelf/32_3_crop_2241_2284_RetractFromShelf.mp4
228 | RetractFromShelf/19_1_crop_238_267_RetractFromShelf.mp4
229 | RetractFromShelf/21_1_crop_3148_3175_RetractFromShelf.mp4
230 | RetractFromShelf/32_3_crop_3332_3375_RetractFromShelf.mp4
231 | RetractFromShelf/19_2_crop_1381_1428_RetractFromShelf.mp4
232 | RetractFromShelf/5_2_crop_3147_3196_RetractFromShelf.mp4
233 | RetractFromShelf/15_2_crop_3307_3337_RetractFromShelf.mp4
234 | RetractFromShelf/16_1_crop_3904_3923_RetractFromShelf.mp4
235 | RetractFromShelf/28_3_crop_1386_1416_RetractFromShelf.mp4
236 | RetractFromShelf/7_2_crop_1900_1920_RetractFromShelf.mp4
237 | RetractFromShelf/40_1_crop_1811_1840_RetractFromShelf.mp4
238 | RetractFromShelf/22_3_crop_1049_1087_RetractFromShelf.mp4
239 | RetractFromShelf/7_2_crop_546_565_RetractFromShelf.mp4
240 | RetractFromShelf/17_3_crop_2392_2459_RetractFromShelf.mp4
241 | RetractFromShelf/7_2_crop_1094_1129_RetractFromShelf.mp4
242 | RetractFromShelf/26_1_crop_2816_2837_RetractFromShelf.mp4
243 | RetractFromShelf/14_2_crop_1102_1129_RetractFromShelf.mp4
244 | RetractFromShelf/7_1_crop_3448_3464_RetractFromShelf.mp4
245 | RetractFromShelf/17_2_crop_4011_4054_RetractFromShelf.mp4
246 | RetractFromShelf/28_2_crop_1474_1531_RetractFromShelf.mp4
247 | RetractFromShelf/12_1_crop_1402_1440_RetractFromShelf.mp4
248 | RetractFromShelf/41_1_crop_388_424_RetractFromShelf.mp4
249 | RetractFromShelf/4_1_crop_2770_2812_RetractFromShelf.mp4
250 | RetractFromShelf/26_1_crop_3010_3035_RetractFromShelf.mp4
251 | RetractFromShelf/26_1_crop_1691_1726_RetractFromShelf.mp4
252 | RetractFromShelf/12_2_crop_691_720_RetractFromShelf.mp4
253 | RetractFromShelf/21_1_crop_3976_3992_RetractFromShelf.mp4
254 | RetractFromShelf/29_3_crop_3646_3662_RetractFromShelf.mp4
255 | RetractFromShelf/8_2_crop_2072_2092_RetractFromShelf.mp4
256 | RetractFromShelf/1_1_crop_2936_2957_RetractFromShelf.mp4
257 | RetractFromShelf/20_1_crop_3142_3168_RetractFromShelf.mp4
258 | RetractFromShelf/26_1_crop_1812_1836_RetractFromShelf.mp4
259 | RetractFromShelf/7_3_crop_3072_3102_RetractFromShelf.mp4
260 | RetractFromShelf/4_3_crop_3027_3067_RetractFromShelf.mp4
261 | RetractFromShelf/14_1_crop_3425_3444_RetractFromShelf.mp4
262 | RetractFromShelf/1_2_crop_3406_3438_RetractFromShelf.mp4
263 | RetractFromShelf/12_1_crop_2812_2846_RetractFromShelf.mp4
264 | RetractFromShelf/19_1_crop_2972_3007_RetractFromShelf.mp4
265 | RetractFromShelf/28_3_crop_2467_2512_RetractFromShelf.mp4
266 | RetractFromShelf/6_2_crop_686_741_RetractFromShelf.mp4
267 | RetractFromShelf/14_3_crop_1542_1583_RetractFromShelf.mp4
268 | RetractFromShelf/6_2_crop_3502_3546_RetractFromShelf.mp4
269 | RetractFromShelf/6_2_crop_2475_2522_RetractFromShelf.mp4
270 | RetractFromShelf/18_2_crop_1094_1153_RetractFromShelf.mp4
271 | RetractFromShelf/5_3_crop_554_579_RetractFromShelf.mp4
272 | RetractFromShelf/14_3_crop_422_461_RetractFromShelf.mp4
273 | RetractFromShelf/11_3_crop_307_347_RetractFromShelf.mp4
274 | RetractFromShelf/6_2_crop_1713_1796_RetractFromShelf.mp4
275 | RetractFromShelf/38_1_crop_1396_1430_RetractFromShelf.mp4
276 | RetractFromShelf/18_2_crop_2225_2246_RetractFromShelf.mp4
277 | RetractFromShelf/7_1_crop_1280_1297_RetractFromShelf.mp4
278 | RetractFromShelf/6_3_crop_852_972_RetractFromShelf.mp4
279 | RetractFromShelf/27_2_crop_3267_3324_RetractFromShelf.mp4
280 | RetractFromShelf/13_1_crop_3684_3700_RetractFromShelf.mp4
281 | RetractFromShelf/32_2_crop_3327_3347_RetractFromShelf.mp4
282 | RetractFromShelf/5_3_crop_1309_1343_RetractFromShelf.mp4
283 | RetractFromShelf/4_3_crop_427_489_RetractFromShelf.mp4
284 | RetractFromShelf/1_1_crop_418_433_RetractFromShelf.mp4
285 | RetractFromShelf/27_2_crop_3052_3105_RetractFromShelf.mp4
286 | RetractFromShelf/33_1_crop_260_295_RetractFromShelf.mp4
287 | RetractFromShelf/27_1_crop_2608_2641_RetractFromShelf.mp4
288 | RetractFromShelf/22_3_crop_1805_1835_RetractFromShelf.mp4
289 | RetractFromShelf/25_3_crop_788_833_RetractFromShelf.mp4
290 | RetractFromShelf/21_1_crop_3877_3908_RetractFromShelf.mp4
291 | RetractFromShelf/4_2_crop_3709_3741_RetractFromShelf.mp4
292 | RetractFromShelf/17_2_crop_1482_1499_RetractFromShelf.mp4
293 | RetractFromShelf/32_1_crop_3203_3242_RetractFromShelf.mp4
294 | RetractFromShelf/24_1_crop_3533_3550_RetractFromShelf.mp4
295 | RetractFromShelf/32_1_crop_3676_3707_RetractFromShelf.mp4
296 | RetractFromShelf/4_1_crop_384_422_RetractFromShelf.mp4
297 | RetractFromShelf/2_1_crop_340_384_RetractFromShelf.mp4
298 | RetractFromShelf/20_2_crop_336_364_RetractFromShelf.mp4
299 | RetractFromShelf/4_1_crop_1718_1758_RetractFromShelf.mp4
300 | RetractFromShelf/18_1_crop_3558_3580_RetractFromShelf.mp4
301 | RetractFromShelf/7_3_crop_2755_2817_RetractFromShelf.mp4
302 | RetractFromShelf/13_1_crop_674_697_RetractFromShelf.mp4
303 | RetractFromShelf/15_2_crop_2299_2322_RetractFromShelf.mp4
304 | RetractFromShelf/26_1_crop_2326_2353_RetractFromShelf.mp4
305 | RetractFromShelf/18_3_crop_1518_1586_RetractFromShelf.mp4
306 | RetractFromShelf/19_1_crop_1065_1107_RetractFromShelf.mp4
307 | RetractFromShelf/32_3_crop_346_412_RetractFromShelf.mp4
308 | RetractFromShelf/30_2_crop_1479_1511_RetractFromShelf.mp4
309 | RetractFromShelf/20_3_crop_2875_2898_RetractFromShelf.mp4
310 | RetractFromShelf/39_1_crop_3092_3116_RetractFromShelf.mp4
311 | RetractFromShelf/2_1_crop_2180_2196_RetractFromShelf.mp4
312 | RetractFromShelf/23_1_crop_813_881_RetractFromShelf.mp4
313 | RetractFromShelf/39_1_crop_2788_2823_RetractFromShelf.mp4
314 | RetractFromShelf/4_2_crop_695_748_RetractFromShelf.mp4
315 | RetractFromShelf/3_3_crop_2337_2361_RetractFromShelf.mp4
316 | RetractFromShelf/15_3_crop_3308_3355_RetractFromShelf.mp4
317 | RetractFromShelf/6_1_crop_500_537_RetractFromShelf.mp4
318 | RetractFromShelf/16_3_crop_3457_3485_RetractFromShelf.mp4
319 | RetractFromShelf/20_2_crop_2933_2973_RetractFromShelf.mp4
320 | RetractFromShelf/15_3_crop_2868_2901_RetractFromShelf.mp4
321 | RetractFromShelf/32_2_crop_3278_3302_RetractFromShelf.mp4
322 | RetractFromShelf/1_3_crop_2471_2492_RetractFromShelf.mp4
323 | RetractFromShelf/32_1_crop_676_733_RetractFromShelf.mp4
324 | RetractFromShelf/10_3_crop_619_634_RetractFromShelf.mp4
325 | RetractFromShelf/18_1_crop_3703_3735_RetractFromShelf.mp4
326 | RetractFromShelf/39_1_crop_3779_3827_RetractFromShelf.mp4
327 | RetractFromShelf/1_2_crop_1909_1923_RetractFromShelf.mp4
328 | RetractFromShelf/7_2_crop_876_903_RetractFromShelf.mp4
329 | RetractFromShelf/17_1_crop_3094_3110_RetractFromShelf.mp4
330 | RetractFromShelf/10_1_crop_2396_2417_RetractFromShelf.mp4
331 | RetractFromShelf/2_1_crop_1522_1546_RetractFromShelf.mp4
332 | RetractFromShelf/2_1_crop_3351_3372_RetractFromShelf.mp4
333 | RetractFromShelf/20_1_crop_3374_3404_RetractFromShelf.mp4
334 | HandInShelf/34_1_crop_635_663_HandInShelf.mp4
335 | HandInShelf/19_2_crop_771_889_HandInShelf.mp4
336 | HandInShelf/8_1_crop_2788_3009_HandInShelf.mp4
337 | HandInShelf/38_1_crop_2459_2529_HandInShelf.mp4
338 | HandInShelf/30_3_crop_2747_2789_HandInShelf.mp4
339 | HandInShelf/16_2_crop_2768_2805_HandInShelf.mp4
340 | HandInShelf/32_2_crop_1645_1671_HandInShelf.mp4
341 | HandInShelf/24_3_crop_3761_3787_HandInShelf.mp4
342 | HandInShelf/39_1_crop_1482_1509_HandInShelf.mp4
343 | HandInShelf/26_2_crop_911_969_HandInShelf.mp4
344 | HandInShelf/2_2_crop_548_575_HandInShelf.mp4
345 | HandInShelf/11_2_crop_2493_2525_HandInShelf.mp4
346 | HandInShelf/19_1_crop_598_693_HandInShelf.mp4
347 | HandInShelf/25_2_crop_606_681_HandInShelf.mp4
348 | HandInShelf/33_1_crop_594_665_HandInShelf.mp4
349 | HandInShelf/26_3_crop_1289_1336_HandInShelf.mp4
350 | HandInShelf/19_3_crop_3020_3127_HandInShelf.mp4
351 | HandInShelf/25_3_crop_1045_1105_HandInShelf.mp4
352 | HandInShelf/2_1_crop_3035_3052_HandInShelf.mp4
353 | HandInShelf/40_1_crop_2683_2919_HandInShelf.mp4
354 | HandInShelf/20_3_crop_1409_1445_HandInShelf.mp4
355 | HandInShelf/28_3_crop_1087_1383_HandInShelf.mp4
356 | HandInShelf/16_2_crop_2629_2659_HandInShelf.mp4
357 | HandInShelf/15_2_crop_1741_1852_HandInShelf.mp4
358 | HandInShelf/1_3_crop_4197_4455_HandInShelf.mp4
359 | HandInShelf/31_2_crop_1118_1150_HandInShelf.mp4
360 | HandInShelf/34_1_crop_558_583_HandInShelf.mp4
361 | HandInShelf/21_2_crop_3181_3241_HandInShelf.mp4
362 | HandInShelf/16_1_crop_437_488_HandInShelf.mp4
363 | HandInShelf/21_2_crop_1712_1836_HandInShelf.mp4
364 | HandInShelf/32_2_crop_894_902_HandInShelf.mp4
365 | HandInShelf/21_2_crop_2225_2529_HandInShelf.mp4
366 | HandInShelf/30_2_crop_3391_3478_HandInShelf.mp4
367 | HandInShelf/8_3_crop_1401_1742_HandInShelf.mp4
368 | HandInShelf/4_3_crop_727_756_HandInShelf.mp4
369 | HandInShelf/12_1_crop_458_532_HandInShelf.mp4
370 | HandInShelf/29_3_crop_1101_1263_HandInShelf.mp4
371 | HandInShelf/1_3_crop_3755_3794_HandInShelf.mp4
372 | HandInShelf/2_2_crop_2452_2505_HandInShelf.mp4
373 | HandInShelf/10_2_crop_477_544_HandInShelf.mp4
374 | HandInShelf/31_3_crop_1493_1523_HandInShelf.mp4
375 | HandInShelf/24_1_crop_1699_1818_HandInShelf.mp4
376 | HandInShelf/16_1_crop_552_573_HandInShelf.mp4
377 | HandInShelf/19_3_crop_277_311_HandInShelf.mp4
378 | HandInShelf/2_1_crop_472_554_HandInShelf.mp4
379 | HandInShelf/16_1_crop_3891_3902_HandInShelf.mp4
380 | HandInShelf/17_1_crop_145_164_HandInShelf.mp4
381 | HandInShelf/19_1_crop_982_1065_HandInShelf.mp4
382 | HandInShelf/30_1_crop_2274_2334_HandInShelf.mp4
383 | HandInShelf/1_1_crop_406_416_HandInShelf.mp4
384 | HandInShelf/7_1_crop_819_918_HandInShelf.mp4
385 | HandInShelf/32_3_crop_1207_1374_HandInShelf.mp4
386 | HandInShelf/23_3_crop_4755_4824_HandInShelf.mp4
387 | HandInShelf/39_1_crop_3525_3587_HandInShelf.mp4
388 | HandInShelf/15_2_crop_728_749_HandInShelf.mp4
389 | HandInShelf/1_2_crop_2352_2374_HandInShelf.mp4
390 | InspectProduct/17_2_crop_1639_1853_InspectProduct.mp4
391 | InspectProduct/31_3_crop_1193_1302_InspectProduct.mp4
392 | InspectProduct/32_3_crop_2617_2883_InspectProduct.mp4
393 | InspectProduct/28_3_crop_237_459_InspectProduct.mp4
394 | InspectProduct/28_2_crop_3956_4142_InspectProduct.mp4
395 | InspectProduct/13_3_crop_3129_3345_InspectProduct.mp4
396 | InspectProduct/1_3_crop_1499_1594_InspectProduct.mp4
397 | InspectProduct/20_3_crop_4196_4248_InspectProduct.mp4
398 | InspectProduct/26_3_crop_605_763_InspectProduct.mp4
399 | InspectProduct/12_1_crop_3320_3515_InspectProduct.mp4
400 | InspectProduct/30_3_crop_2825_3459_InspectProduct.mp4
401 | InspectProduct/12_3_crop_2732_2760_InspectProduct.mp4
402 | InspectProduct/5_2_crop_3199_3357_InspectProduct.mp4
403 | InspectProduct/12_2_crop_1692_1763_InspectProduct.mp4
404 | InspectProduct/20_3_crop_429_551_InspectProduct.mp4
405 | InspectProduct/14_2_crop_1367_2016_InspectProduct.mp4
406 | InspectProduct/25_3_crop_834_851_InspectProduct.mp4
407 | InspectProduct/39_1_crop_2419_2457_InspectProduct.mp4
408 | InspectProduct/1_3_crop_1253_1382_InspectProduct.mp4
409 | InspectProduct/28_3_crop_2517_2534_InspectProduct.mp4
410 | InspectProduct/17_1_crop_2737_2903_InspectProduct.mp4
411 | InspectProduct/20_1_crop_2770_2797_InspectProduct.mp4
412 | InspectProduct/22_3_crop_569_683_InspectProduct.mp4
413 | InspectProduct/7_3_crop_2405_2431_InspectProduct.mp4
414 | InspectProduct/19_2_crop_2584_2619_InspectProduct.mp4
415 | InspectProduct/34_1_crop_3577_3717_InspectProduct.mp4
416 | InspectProduct/17_3_crop_1534_1585_InspectProduct.mp4
417 | InspectProduct/11_1_crop_809_1003_InspectProduct.mp4
418 | InspectProduct/28_1_crop_2957_3012_InspectProduct.mp4
419 | InspectProduct/12_2_crop_691_719_InspectProduct.mp4
420 | InspectProduct/32_2_crop_3614_3641_InspectProduct.mp4
421 | InspectProduct/31_3_crop_2268_2346_InspectProduct.mp4
422 | InspectProduct/21_3_crop_2629_2784_InspectProduct.mp4
423 | InspectProduct/4_2_crop_3554_3667_InspectProduct.mp4
424 | InspectProduct/10_2_crop_1357_1512_InspectProduct.mp4
425 | InspectProduct/10_3_crop_3110_3177_InspectProduct.mp4
426 | InspectProduct/25_1_crop_2365_2526_InspectProduct.mp4
427 | InspectProduct/20_1_crop_3201_3233_InspectProduct.mp4
428 | InspectProduct/17_1_crop_2091_2152_InspectProduct.mp4
429 | InspectProduct/29_2_crop_3015_3039_InspectProduct.mp4
430 | InspectProduct/3_3_crop_1629_1645_InspectProduct.mp4
431 | InspectProduct/37_1_crop_3741_3793_InspectProduct.mp4
432 | InspectProduct/9_3_crop_2729_2848_InspectProduct.mp4
433 | InspectProduct/3_1_crop_288_368_InspectProduct.mp4
434 | InspectProduct/7_1_crop_1184_1237_InspectProduct.mp4
435 | InspectProduct/16_1_crop_2365_2404_InspectProduct.mp4
436 | InspectProduct/15_2_crop_204_228_InspectProduct.mp4
437 | InspectProduct/5_2_crop_1114_1279_InspectProduct.mp4
438 | InspectProduct/34_1_crop_3306_3384_InspectProduct.mp4
439 | InspectProduct/15_3_crop_733_810_InspectProduct.mp4
440 | InspectProduct/22_1_crop_1124_1444_InspectProduct.mp4
441 | InspectProduct/26_2_crop_3553_3567_InspectProduct.mp4
442 | InspectProduct/13_1_crop_3473_3648_InspectProduct.mp4
443 | InspectProduct/17_2_crop_3315_3379_InspectProduct.mp4
444 | InspectProduct/7_2_crop_3439_3492_InspectProduct.mp4
445 | InspectProduct/35_1_crop_2831_2848_InspectProduct.mp4
446 | InspectProduct/14_1_crop_495_598_InspectProduct.mp4
447 | InspectProduct/30_2_crop_767_1432_InspectProduct.mp4
448 | InspectProduct/1_3_crop_2495_2640_InspectProduct.mp4
449 | InspectProduct/28_2_crop_3659_3858_InspectProduct.mp4
450 | InspectProduct/1_3_crop_2767_2784_InspectProduct.mp4
451 | InspectProduct/6_1_crop_3150_3211_InspectProduct.mp4
452 | InspectProduct/7_2_crop_3670_3704_InspectProduct.mp4
453 | InspectProduct/3_2_crop_1061_1096_InspectProduct.mp4
454 | InspectProduct/21_1_crop_2854_2919_InspectProduct.mp4
455 | InspectProduct/1_3_crop_667_829_InspectProduct.mp4
456 | InspectProduct/23_2_crop_2984_3032_InspectProduct.mp4
457 | InspectShelf/16_2_crop_3280_3426_InspectShelf.mp4
458 | InspectShelf/1_1_crop_3458_3575_InspectShelf.mp4
459 | InspectShelf/26_2_crop_1050_1161_InspectShelf.mp4
460 | InspectShelf/22_2_crop_1015_1173_InspectShelf.mp4
461 | InspectShelf/22_3_crop_1090_1252_InspectShelf.mp4
462 | InspectShelf/17_3_crop_3160_3264_InspectShelf.mp4
463 | InspectShelf/5_3_crop_582_615_InspectShelf.mp4
464 | InspectShelf/41_2_crop_1447_1523_InspectShelf.mp4
465 | InspectShelf/6_2_crop_3171_3461_InspectShelf.mp4
466 | InspectShelf/36_1_crop_821_896_InspectShelf.mp4
467 | InspectShelf/22_1_crop_3688_3827_InspectShelf.mp4
468 | InspectShelf/31_1_crop_3463_3518_InspectShelf.mp4
469 | InspectShelf/31_2_crop_1892_1955_InspectShelf.mp4
470 | InspectShelf/24_1_crop_890_1058_InspectShelf.mp4
471 | InspectShelf/32_1_crop_239_327_InspectShelf.mp4
472 | InspectShelf/39_1_crop_2997_3049_InspectShelf.mp4
473 | InspectShelf/5_1_crop_836_1037_InspectShelf.mp4
474 | InspectShelf/12_3_crop_749_836_InspectShelf.mp4
475 | InspectShelf/7_1_crop_1778_1889_InspectShelf.mp4
476 | InspectShelf/6_3_crop_1233_1274_InspectShelf.mp4
477 | InspectShelf/2_1_crop_1947_2057_InspectShelf.mp4
478 | InspectShelf/22_2_crop_460_585_InspectShelf.mp4
479 | InspectShelf/13_2_crop_2889_2932_InspectShelf.mp4
480 | InspectShelf/31_2_crop_1248_1470_InspectShelf.mp4
481 | InspectShelf/10_2_crop_579_615_InspectShelf.mp4
482 | InspectShelf/7_2_crop_1045_1055_InspectShelf.mp4
483 | InspectShelf/4_2_crop_849_1034_InspectShelf.mp4
484 | InspectShelf/39_1_crop_2460_2733_InspectShelf.mp4
485 | InspectShelf/17_3_crop_1189_1223_InspectShelf.mp4
486 | InspectShelf/12_3_crop_3296_3423_InspectShelf.mp4
487 | InspectShelf/11_3_crop_2815_3081_InspectShelf.mp4
488 | InspectShelf/12_3_crop_554_584_InspectShelf.mp4
489 | InspectShelf/2_2_crop_2161_2243_InspectShelf.mp4
490 | InspectShelf/22_3_crop_1955_1979_InspectShelf.mp4
491 | InspectShelf/5_2_crop_398_520_InspectShelf.mp4
492 | InspectShelf/29_1_crop_265_800_InspectShelf.mp4
493 | InspectShelf/36_1_crop_3187_3299_InspectShelf.mp4
494 | InspectShelf/26_1_crop_3329_3410_InspectShelf.mp4
495 | InspectShelf/2_2_crop_2755_2875_InspectShelf.mp4
496 | InspectShelf/38_1_crop_3602_3727_InspectShelf.mp4
497 | InspectShelf/32_2_crop_3675_3744_InspectShelf.mp4
498 | InspectShelf/12_2_crop_2591_2715_InspectShelf.mp4
499 | InspectShelf/19_2_crop_2622_2681_InspectShelf.mp4
500 | InspectShelf/4_2_crop_2869_3280_InspectShelf.mp4
501 | InspectShelf/20_1_crop_2654_2709_InspectShelf.mp4
502 | InspectShelf/28_2_crop_2463_2586_InspectShelf.mp4
503 | InspectShelf/17_3_crop_3129_3138_InspectShelf.mp4
504 | InspectShelf/13_3_crop_2801_3055_InspectShelf.mp4
505 | InspectShelf/12_2_crop_541_581_InspectShelf.mp4
506 | InspectShelf/24_3_crop_547_666_InspectShelf.mp4
507 | InspectShelf/4_3_crop_2902_2993_InspectShelf.mp4
508 | InspectShelf/27_2_crop_1072_1150_InspectShelf.mp4
509 | InspectShelf/33_1_crop_3635_3700_InspectShelf.mp4
510 | InspectShelf/27_2_crop_3413_3445_InspectShelf.mp4
511 | InspectShelf/10_1_crop_865_965_InspectShelf.mp4
512 | InspectShelf/2_3_crop_2219_2258_InspectShelf.mp4
513 | InspectShelf/8_3_crop_3630_3667_InspectShelf.mp4
514 | InspectShelf/15_1_crop_169_1208_InspectShelf.mp4
515 | InspectShelf/1_3_crop_580_611_InspectShelf.mp4
516 | InspectShelf/31_2_crop_2581_2639_InspectShelf.mp4
517 | InspectShelf/3_1_crop_1344_1585_InspectShelf.mp4
518 | InspectShelf/39_1_crop_862_1023_InspectShelf.mp4
519 | InspectShelf/4_2_crop_1143_1716_InspectShelf.mp4
520 | InspectShelf/25_2_crop_3877_4245_InspectShelf.mp4
521 | InspectShelf/22_2_crop_3063_3227_InspectShelf.mp4
522 | InspectShelf/4_3_crop_885_1015_InspectShelf.mp4
523 | InspectShelf/9_2_crop_1419_1557_InspectShelf.mp4
524 | InspectShelf/28_2_crop_3234_3507_InspectShelf.mp4
525 | InspectShelf/22_3_crop_3136_3226_InspectShelf.mp4
526 | InspectShelf/41_1_crop_565_651_InspectShelf.mp4
527 | InspectShelf/31_1_crop_1028_1233_InspectShelf.mp4
528 | InspectShelf/26_2_crop_3088_3256_InspectShelf.mp4
529 | InspectShelf/30_3_crop_2510_2578_InspectShelf.mp4
530 | InspectShelf/25_1_crop_232_264_InspectShelf.mp4
531 | InspectShelf/7_3_crop_2631_2716_InspectShelf.mp4
532 | InspectShelf/41_2_crop_2423_2532_InspectShelf.mp4
533 | InspectShelf/29_3_crop_4094_4343_InspectShelf.mp4
534 | InspectShelf/26_3_crop_1397_1885_InspectShelf.mp4
535 | InspectShelf/3_3_crop_2874_3078_InspectShelf.mp4
536 | InspectShelf/18_2_crop_3178_3366_InspectShelf.mp4
537 | 


--------------------------------------------------------------------------------
/data/movefile.py:
--------------------------------------------------------------------------------
 1 | """
 2 | After extracting the RAR, we run this to move all the files into
 3 | the appropriate train/test folders.
 4 | 
 5 | Should only run this file once!
 6 | """
 7 | import os
 8 | import os.path
 9 | 
10 | def get_train_test_lists(version='01'):
11 |     """
12 |     Using one of the train/test files (01, 02, or 03), get the filename
13 |     breakdowns we'll later use to move everything.
14 |     """
15 |     # Get our files based on version.
16 |     test_file = './ucfTrainTestlist/testlist' + version + '.txt'
17 |     train_file = './ucfTrainTestlist/trainlist' + version + '.txt'
18 |     val_file = './ucfTrainTestlist/validationlist' + version + '.txt'
19 |     # Build the test list.
20 |     with open(test_file) as fin:
21 |         test_list = [row.strip() for row in list(fin)]
22 | 
23 |     # Build the train list. Extra step to remove the class index.
24 |     with open(train_file) as fin:
25 |         train_list = [row.strip() for row in list(fin)]
26 |         train_list = [row.split(' ')[0] for row in train_list]
27 | 
28 |     with open(val_file) as fin:
29 |         val_list = [row.strip() for row in list(fin)]
30 |         val_list = [row.split(' ')[0] for row in val_list]
31 | 
32 |     # Set the groups in a dictionary.
33 |     file_groups = {
34 |         'train': train_list,
35 |         'test': test_list,
36 |         'validation': val_list
37 |     }
38 | 
39 |     return file_groups
40 | 
41 | def move_files(file_groups):
42 |     """This assumes all of our files are currently in _this_ directory.
43 |     So move them to the appropriate spot. Only needs to happen once.
44 |     """
45 |     # Do each of our groups.
46 |     for group, videos in file_groups.items():
47 | 
48 |         # Do each of our videos.
49 |         for video in videos:
50 | 
51 |             # Get the parts.
52 |             parts = video.split('/')
53 |             classname = parts[0]
54 |             filename = parts[1]
55 | 
56 |             # Check if this class exists.
57 |             if not os.path.exists(group + '/' + classname):
58 |                 print("Creating folder for %s/%s" % (group, classname))
59 |                 os.makedirs(group + '/' + classname)
60 | 
61 |             # Check if we have already moved this file, or at least that it
62 |             # exists to move.
63 |             if not os.path.exists(filename):
64 |                 print("Can't find %s to move. Skipping." % (filename))
65 |                 continue
66 | 
67 |             # Move it.
68 |             dest = group + '/' + classname + '/' + filename
69 |             print("Moving %s to %s" % (filename, dest))
70 |             os.rename(filename, dest)
71 | 
72 |     print("Done.")
73 | 
74 | def main():
75 |     """
76 |     Go through each of our train/test text files and move the videos
77 |     to the right place.
78 |     """
79 |     # Get the videos in groups so we can move them.
80 |     group_lists = get_train_test_lists()
81 | 
82 |     # Move the files.
83 |     move_files(group_lists)
84 | 
85 | if __name__ == '__main__':
86 |     main()


--------------------------------------------------------------------------------
/data/ucfTrainTestlist/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/naviocean/pseudo-3d-pytorch/50297d11248630792709782f467982e80c281384/data/ucfTrainTestlist/.DS_Store


--------------------------------------------------------------------------------
/data/ucfTrainTestlist/classInd.txt:
--------------------------------------------------------------------------------
  1 | 1 ApplyEyeMakeup
  2 | 2 ApplyLipstick
  3 | 3 Archery
  4 | 4 BabyCrawling
  5 | 5 BalanceBeam
  6 | 6 BandMarching
  7 | 7 BaseballPitch
  8 | 8 Basketball
  9 | 9 BasketballDunk
 10 | 10 BenchPress
 11 | 11 Biking
 12 | 12 Billiards
 13 | 13 BlowDryHair
 14 | 14 BlowingCandles
 15 | 15 BodyWeightSquats
 16 | 16 Bowling
 17 | 17 BoxingPunchingBag
 18 | 18 BoxingSpeedBag
 19 | 19 BreastStroke
 20 | 20 BrushingTeeth
 21 | 21 CleanAndJerk
 22 | 22 CliffDiving
 23 | 23 CricketBowling
 24 | 24 CricketShot
 25 | 25 CuttingInKitchen
 26 | 26 Diving
 27 | 27 Drumming
 28 | 28 Fencing
 29 | 29 FieldHockeyPenalty
 30 | 30 FloorGymnastics
 31 | 31 FrisbeeCatch
 32 | 32 FrontCrawl
 33 | 33 GolfSwing
 34 | 34 Haircut
 35 | 35 Hammering
 36 | 36 HammerThrow
 37 | 37 HandstandPushups
 38 | 38 HandstandWalking
 39 | 39 HeadMassage
 40 | 40 HighJump
 41 | 41 HorseRace
 42 | 42 HorseRiding
 43 | 43 HulaHoop
 44 | 44 IceDancing
 45 | 45 JavelinThrow
 46 | 46 JugglingBalls
 47 | 47 JumpingJack
 48 | 48 JumpRope
 49 | 49 Kayaking
 50 | 50 Knitting
 51 | 51 LongJump
 52 | 52 Lunges
 53 | 53 MilitaryParade
 54 | 54 Mixing
 55 | 55 MoppingFloor
 56 | 56 Nunchucks
 57 | 57 ParallelBars
 58 | 58 PizzaTossing
 59 | 59 PlayingCello
 60 | 60 PlayingDaf
 61 | 61 PlayingDhol
 62 | 62 PlayingFlute
 63 | 63 PlayingGuitar
 64 | 64 PlayingPiano
 65 | 65 PlayingSitar
 66 | 66 PlayingTabla
 67 | 67 PlayingViolin
 68 | 68 PoleVault
 69 | 69 PommelHorse
 70 | 70 PullUps
 71 | 71 Punch
 72 | 72 PushUps
 73 | 73 Rafting
 74 | 74 RockClimbingIndoor
 75 | 75 RopeClimbing
 76 | 76 Rowing
 77 | 77 SalsaSpin
 78 | 78 ShavingBeard
 79 | 79 Shotput
 80 | 80 SkateBoarding
 81 | 81 Skiing
 82 | 82 Skijet
 83 | 83 SkyDiving
 84 | 84 SoccerJuggling
 85 | 85 SoccerPenalty
 86 | 86 StillRings
 87 | 87 SumoWrestling
 88 | 88 Surfing
 89 | 89 Swing
 90 | 90 TableTennisShot
 91 | 91 TaiChi
 92 | 92 TennisSwing
 93 | 93 ThrowDiscus
 94 | 94 TrampolineJumping
 95 | 95 Typing
 96 | 96 UnevenBars
 97 | 97 VolleyballSpiking
 98 | 98 WalkingWithDog
 99 | 99 WallPushups
100 | 100 WritingOnBoard
101 | 101 YoYo
102 | 


--------------------------------------------------------------------------------
/dog.0.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/naviocean/pseudo-3d-pytorch/50297d11248630792709782f467982e80c281384/dog.0.jpg


--------------------------------------------------------------------------------
/logger.py:
--------------------------------------------------------------------------------
 1 | import logging
 2 | 
 3 | formatter = logging.Formatter('%(asctime)s %(levelname)s %(message)s')
 4 | 
 5 | 
 6 | def Logger(name, log_file, level=logging.INFO):
 7 |     """Function setup as many loggers as you want"""
 8 | 
 9 |     handler = logging.FileHandler(log_file)
10 |     handler.setFormatter(formatter)
11 | 
12 |     logger = logging.getLogger(name)
13 |     logger.setLevel(level)
14 |     logger.addHandler(handler)
15 | 
16 |     return logger
17 | 


--------------------------------------------------------------------------------
/lr_scheduler.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | from torch.optim import Optimizer
  3 | 
  4 | 
  5 | class CyclicLR(object):
  6 |     """Sets the learning rate of each parameter group according to
  7 |     cyclical learning rate policy (CLR). The policy cycles the learning
  8 |     rate between two boundaries with a constant frequency, as detailed in
  9 |     the paper `Cyclical Learning Rates for Training Neural Networks`_.
 10 |     The distance between the two boundaries can be scaled on a per-iteration
 11 |     or per-cycle basis.
 12 |     Cyclical learning rate policy changes the learning rate after every batch.
 13 |     `batch_step` should be called after a batch has been used for training.
 14 |     To resume training, save `last_batch_iteration` and use it to instantiate `CycleLR`.
 15 |     This class has three built-in policies, as put forth in the paper:
 16 |     "triangular":
 17 |         A basic triangular cycle w/ no amplitude scaling.
 18 |     "triangular2":
 19 |         A basic triangular cycle that scales initial amplitude by half each cycle.
 20 |     "exp_range":
 21 |         A cycle that scales initial amplitude by gamma**(cycle iterations) at each
 22 |         cycle iteration.
 23 |     This implementation was adapted from the github repo: `bckenstler/CLR`_
 24 |     Args:
 25 |         optimizer (Optimizer): Wrapped optimizer.
 26 |         base_lr (float or list): Initial learning rate which is the
 27 |             lower boundary in the cycle for eachparam groups.
 28 |             Default: 0.001
 29 |         max_lr (float or list): Upper boundaries in the cycle for
 30 |             each parameter group. Functionally,
 31 |             it defines the cycle amplitude (max_lr - base_lr).
 32 |             The lr at any cycle is the sum of base_lr
 33 |             and some scaling of the amplitude; therefore
 34 |             max_lr may not actually be reached depending on
 35 |             scaling function. Default: 0.006
 36 |         step_size (int): Number of training iterations per
 37 |             half cycle. Authors suggest setting step_size
 38 |             2-8 x training iterations in epoch. Default: 2000
 39 |         mode (str): One of {triangular, triangular2, exp_range}.
 40 |             Values correspond to policies detailed above.
 41 |             If scale_fn is not None, this argument is ignored.
 42 |             Default: 'triangular'
 43 |         gamma (float): Constant in 'exp_range' scaling function:
 44 |             gamma**(cycle iterations)
 45 |             Default: 1.0
 46 |         scale_fn (function): Custom scaling policy defined by a single
 47 |             argument lambda function, where
 48 |             0 <= scale_fn(x) <= 1 for all x >= 0.
 49 |             mode paramater is ignored
 50 |             Default: None
 51 |         scale_mode (str): {'cycle', 'iterations'}.
 52 |             Defines whether scale_fn is evaluated on
 53 |             cycle number or cycle iterations (training
 54 |             iterations since start of cycle).
 55 |             Default: 'cycle'
 56 |         last_batch_iteration (int): The index of the last batch. Default: -1
 57 |     Example:
 58 |         >>> optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.9)
 59 |         >>> scheduler = torch.optim.CyclicLR(optimizer)
 60 |         >>> data_loader = torch.utils.data.DataLoader(...)
 61 |         >>> for epoch in range(10):
 62 |         >>>     for batch in data_loader:
 63 |         >>>         scheduler.batch_step()
 64 |         >>>         train_batch(...)
 65 |     .. _Cyclical Learning Rates for Training Neural Networks: https://arxiv.org/abs/1506.01186
 66 |     .. _bckenstler/CLR: https://github.com/bckenstler/CLR
 67 |     """
 68 | 
 69 |     def __init__(self, optimizer, base_lr=1e-3, max_lr=6e-3,
 70 |                  step_size=2000, mode='triangular', gamma=1.,
 71 |                  scale_fn=None, scale_mode='cycle', last_batch_iteration=-1):
 72 | 
 73 |         if not isinstance(optimizer, Optimizer):
 74 |             raise TypeError('{} is not an Optimizer'.format(
 75 |                 type(optimizer).__name__))
 76 |         self.optimizer = optimizer
 77 | 
 78 |         if isinstance(base_lr, list) or isinstance(base_lr, tuple):
 79 |             if len(base_lr) != len(optimizer.param_groups):
 80 |                 raise ValueError("expected {} base_lr, got {}".format(
 81 |                     len(optimizer.param_groups), len(base_lr)))
 82 |             self.base_lrs = list(base_lr)
 83 |         else:
 84 |             self.base_lrs = [base_lr] * len(optimizer.param_groups)
 85 | 
 86 |         if isinstance(max_lr, list) or isinstance(max_lr, tuple):
 87 |             if len(max_lr) != len(optimizer.param_groups):
 88 |                 raise ValueError("expected {} max_lr, got {}".format(
 89 |                     len(optimizer.param_groups), len(max_lr)))
 90 |             self.max_lrs = list(max_lr)
 91 |         else:
 92 |             self.max_lrs = [max_lr] * len(optimizer.param_groups)
 93 | 
 94 |         self.step_size = step_size
 95 | 
 96 |         if mode not in ['triangular', 'triangular2', 'exp_range'] \
 97 |                 and scale_fn is None:
 98 |             raise ValueError('mode is invalid and scale_fn is None')
 99 | 
100 |         self.mode = mode
101 |         self.gamma = gamma
102 | 
103 |         if scale_fn is None:
104 |             if self.mode == 'triangular':
105 |                 self.scale_fn = self._triangular_scale_fn
106 |                 self.scale_mode = 'cycle'
107 |             elif self.mode == 'triangular2':
108 |                 self.scale_fn = self._triangular2_scale_fn
109 |                 self.scale_mode = 'cycle'
110 |             elif self.mode == 'exp_range':
111 |                 self.scale_fn = self._exp_range_scale_fn
112 |                 self.scale_mode = 'iterations'
113 |         else:
114 |             self.scale_fn = scale_fn
115 |             self.scale_mode = scale_mode
116 | 
117 |         self.batch_step(last_batch_iteration + 1)
118 |         self.last_batch_iteration = last_batch_iteration
119 | 
120 |     def batch_step(self, batch_iteration=None):
121 |         if batch_iteration is None:
122 |             batch_iteration = self.last_batch_iteration + 1
123 |         self.last_batch_iteration = batch_iteration
124 |         for param_group, lr in zip(self.optimizer.param_groups, self.get_lr()):
125 |             param_group['lr'] = lr
126 | 
127 |     def _triangular_scale_fn(self, x):
128 |         return 1.
129 | 
130 |     def _triangular2_scale_fn(self, x):
131 |         return 1 / (2. ** (x - 1))
132 | 
133 |     def _exp_range_scale_fn(self, x):
134 |         return self.gamma ** (x)
135 | 
136 |     def get_lr(self):
137 |         step_size = float(self.step_size)
138 |         cycle = np.floor(1 + self.last_batch_iteration / (2 * step_size))
139 |         x = np.abs(self.last_batch_iteration / step_size - 2 * cycle + 1)
140 | 
141 |         lrs = []
142 |         param_lrs = zip(self.optimizer.param_groups, self.base_lrs, self.max_lrs)
143 |         for param_group, base_lr, max_lr in param_lrs:
144 |             base_height = (max_lr - base_lr) * np.maximum(0, (1 - x))
145 |             if self.scale_mode == 'cycle':
146 |                 lr = base_lr + base_height * self.scale_fn(cycle)
147 |             else:
148 |                 lr = base_lr + base_height * self.scale_fn(self.last_batch_iteration)
149 |             lrs.append(lr)
150 |         return lrs
151 | 


--------------------------------------------------------------------------------
/main.py:
--------------------------------------------------------------------------------
 1 | from __future__ import print_function
 2 | import argparse
 3 | from train import Training
 4 | from test import Testing
 5 | 
 6 | parser = argparse.ArgumentParser(description='PyTorch Pseudo-3D fine-tuning')
 7 | parser.add_argument('data', metavar='DIR', help='path to dataset')
 8 | parser.add_argument('--data-set', default='UCF101', const='UCF101', nargs='?', choices=['UCF101', 'Breakfast', 'merl'])
 9 | parser.add_argument('--workers', default=4, type=int, metavar='N', help='number of data loading workers (default: 4)')
10 | parser.add_argument('--early-stop', default=10, type=int, metavar='N', help='number of early stopping')
11 | parser.add_argument('--epochs', default=75, type=int, metavar='N', help='number of total epochs to run')
12 | parser.add_argument('--start-epoch', default=0, type=int, metavar='N', help='manual epoch number (useful on restarts)')
13 | parser.add_argument('-b', '--batch-size', default=16, type=int, metavar='N', help='mini-batch size (default: 256)')
14 | parser.add_argument('--lr', '--learning-rate', default=1e-4, type=float, metavar='LR', help='initial learning rate')
15 | parser.add_argument('--momentum', default=0.9, type=float, metavar='M', help='momentum')
16 | parser.add_argument('--dropout', default=0.5, type=float, metavar='M', help='dropout')
17 | parser.add_argument('--weight-decay', default=1e-4, type=float, metavar='W', help='weight decay')
18 | parser.add_argument('--print-freq', default=1, type=int, metavar='N', help='print frequency')
19 | parser.add_argument('--resume', default='', type=str, metavar='PATH', help='path to latest checkpoint')
20 | parser.add_argument('--evaluate', dest='evaluate', action='store_true', help='evaluate model on validation set')
21 | parser.add_argument('--test', dest='test', action='store_true', help='evaluate model on test set')
22 | parser.add_argument('--random', dest='random', action='store_true', help='random pick image')
23 | parser.add_argument('--pretrained', dest='pretrained', action='store_true', help='use pre-trained model')
24 | parser.add_argument('--model-type', default='P3D', choices=['P3D', 'C3D', 'I3D'], help='which model to run the code')
25 | parser.add_argument('--num-frames', default=16, type=int, metavar='N', help='number frames per clip')
26 | parser.add_argument('--log-visualize', default='./runs', type=str, metavar='PATH', help='tensorboard log')
27 | 
28 | def main():
29 |     args = parser.parse_args()
30 |     args = vars(args)
31 | 
32 |     if args['data_set'] == 'UCF101':
33 |         print('UCF101 data set')
34 |         name_list = 'ucfTrainTestlist'
35 |         num_classes = 101
36 |     elif args['data_set'] == 'Breakfast':
37 |         print("breakfast data set")
38 |         num_classes = 37
39 |         name_list = 'breakfastTrainTestList'
40 |     else:
41 |         print('Merl data set')
42 |         num_classes = 5
43 |         name_list = 'merlTrainTestList'
44 | 
45 |     if args['test']:
46 |         Testing(name_list=name_list, num_classes=num_classes, modality='RGB', **args)
47 |     else:
48 |         Training(name_list=name_list, num_classes=num_classes, modality='RGB', **args)
49 | 
50 | 
51 | if __name__ == '__main__':
52 |     main()
53 | 


--------------------------------------------------------------------------------
/meter.py:
--------------------------------------------------------------------------------
 1 | class AverageMeter(object):
 2 |     """Computes and stores the average and current value"""
 3 | 
 4 |     def __init__(self):
 5 |         self.reset()
 6 | 
 7 |     def reset(self):
 8 |         self.val = 0
 9 |         self.avg = 0
10 |         self.sum = 0
11 |         self.count = 0
12 | 
13 |     def update(self, val, n=1):
14 |         self.val = val
15 |         self.sum += val * n
16 |         self.count += n
17 |         self.avg = self.sum / self.count


--------------------------------------------------------------------------------
/models/C3D.py:
--------------------------------------------------------------------------------
 1 | import torch.nn as nn
 2 | 
 3 | 
 4 | class C3D(nn.Module):
 5 |     """
 6 |     The C3D network as described in [1].
 7 |     """
 8 | 
 9 |     def __init__(self):
10 |         super(C3D, self).__init__()
11 | 
12 |         self.conv1 = nn.Conv3d(3, 64, kernel_size=(3, 3, 3), padding=(1, 1, 1))
13 |         self.pool1 = nn.MaxPool3d(kernel_size=(1, 2, 2), stride=(1, 2, 2))
14 | 
15 |         self.conv2 = nn.Conv3d(64, 128, kernel_size=(3, 3, 3), padding=(1, 1, 1))
16 |         self.pool2 = nn.MaxPool3d(kernel_size=(2, 2, 2), stride=(2, 2, 2))
17 | 
18 |         self.conv3a = nn.Conv3d(128, 256, kernel_size=(3, 3, 3), padding=(1, 1, 1))
19 |         self.conv3b = nn.Conv3d(256, 256, kernel_size=(3, 3, 3), padding=(1, 1, 1))
20 |         self.pool3 = nn.MaxPool3d(kernel_size=(2, 2, 2), stride=(2, 2, 2))
21 | 
22 |         self.conv4a = nn.Conv3d(256, 512, kernel_size=(3, 3, 3), padding=(1, 1, 1))
23 |         self.conv4b = nn.Conv3d(512, 512, kernel_size=(3, 3, 3), padding=(1, 1, 1))
24 |         self.pool4 = nn.MaxPool3d(kernel_size=(2, 2, 2), stride=(2, 2, 2))
25 | 
26 |         self.conv5a = nn.Conv3d(512, 512, kernel_size=(3, 3, 3), padding=(1, 1, 1))
27 |         self.conv5b = nn.Conv3d(512, 512, kernel_size=(3, 3, 3), padding=(1, 1, 1))
28 |         self.pool5 = nn.MaxPool3d(kernel_size=(2, 2, 2), stride=(2, 2, 2), padding=(0, 1, 1))
29 | 
30 |         self.fc6 = nn.Linear(8192, 4096)
31 |         self.fc7 = nn.Linear(4096, 4096)
32 |         self.fc8 = nn.Linear(4096, 487)
33 | 
34 |         self.dropout = nn.Dropout(p=0.5)
35 | 
36 |         self.relu = nn.ReLU()
37 |         self.softmax = nn.Softmax()
38 | 
39 |     def forward(self, x):
40 | 
41 |         h = self.relu(self.conv1(x))
42 |         h = self.pool1(h)
43 | 
44 |         h = self.relu(self.conv2(h))
45 |         h = self.pool2(h)
46 | 
47 |         h = self.relu(self.conv3a(h))
48 |         h = self.relu(self.conv3b(h))
49 |         h = self.pool3(h)
50 | 
51 |         h = self.relu(self.conv4a(h))
52 |         h = self.relu(self.conv4b(h))
53 |         h = self.pool4(h)
54 | 
55 |         h = self.relu(self.conv5a(h))
56 |         h = self.relu(self.conv5b(h))
57 |         h = self.pool5(h)
58 | 
59 |         h = h.view(-1, 8192)
60 |         h = self.relu(self.fc6(h))
61 |         h = self.dropout(h)
62 |         h = self.relu(self.fc7(h))
63 |         h = self.dropout(h)
64 | 
65 |         logits = self.fc8(h)
66 |         probs = self.softmax(logits)
67 | 
68 |         return probs
69 | 
70 | """
71 | References
72 | ----------
73 | [1] Tran, Du, et al. "Learning spatiotemporal features with 3d convolutional networks." 
74 | Proceedings of the IEEE international conference on computer vision. 2015.
75 | """


--------------------------------------------------------------------------------
/models/i3d.py:
--------------------------------------------------------------------------------
  1 | import torch.nn as nn
  2 | import torch
  3 | 
  4 | class BasicConv3d(nn.Module):
  5 |     def __init__(self, in_planes, out_planes, kernel_size, stride, padding=0):
  6 |         super(BasicConv3d, self).__init__()
  7 |         self.conv = nn.Conv3d(in_planes, out_planes, kernel_size=kernel_size, stride=stride, padding=padding,
  8 |                               bias=False)  # verify bias false
  9 | 
 10 |         # verify defalt value in sonnet
 11 |         self.bn = nn.BatchNorm3d(out_planes, eps=1e-3, momentum=0.001, affine=True)
 12 |         self.relu = nn.ReLU(inplace=True)
 13 | 
 14 |     def forward(self, x):
 15 |         x = self.conv(x)
 16 |         x = self.bn(x)
 17 |         x = self.relu(x)
 18 |         return x
 19 | 
 20 | 
 21 | class Mixed_3b(nn.Module):
 22 |     def __init__(self):
 23 |         super(Mixed_3b, self).__init__()
 24 | 
 25 |         self.branch0 = nn.Sequential(
 26 |             BasicConv3d(192, 64, kernel_size=1, stride=1),
 27 |         )
 28 |         self.branch1 = nn.Sequential(
 29 |             BasicConv3d(192, 96, kernel_size=1, stride=1),
 30 |             BasicConv3d(96, 128, kernel_size=3, stride=1, padding=1),
 31 |         )
 32 |         self.branch2 = nn.Sequential(
 33 |             BasicConv3d(192, 16, kernel_size=1, stride=1),
 34 |             BasicConv3d(16, 32, kernel_size=3, stride=1, padding=1),
 35 |         )
 36 |         self.branch3 = nn.Sequential(
 37 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
 38 |             BasicConv3d(192, 32, kernel_size=1, stride=1),
 39 |         )
 40 | 
 41 |     def forward(self, x):
 42 |         x0 = self.branch0(x)
 43 |         x1 = self.branch1(x)
 44 |         x2 = self.branch2(x)
 45 |         x3 = self.branch3(x)
 46 |         out = torch.cat((x0, x1, x2, x3), 1)
 47 | 
 48 |         return out
 49 | 
 50 | 
 51 | class Mixed_3c(nn.Module):
 52 |     def __init__(self):
 53 |         super(Mixed_3c, self).__init__()
 54 |         self.branch0 = nn.Sequential(
 55 |             BasicConv3d(256, 128, kernel_size=1, stride=1),
 56 |         )
 57 |         self.branch1 = nn.Sequential(
 58 |             BasicConv3d(256, 128, kernel_size=1, stride=1),
 59 |             BasicConv3d(128, 192, kernel_size=3, stride=1, padding=1),
 60 |         )
 61 |         self.branch2 = nn.Sequential(
 62 |             BasicConv3d(256, 32, kernel_size=1, stride=1),
 63 |             BasicConv3d(32, 96, kernel_size=3, stride=1, padding=1),
 64 |         )
 65 |         self.branch3 = nn.Sequential(
 66 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
 67 |             BasicConv3d(256, 64, kernel_size=1, stride=1),
 68 |         )
 69 | 
 70 |     def forward(self, x):
 71 |         x0 = self.branch0(x)
 72 |         x1 = self.branch1(x)
 73 |         x2 = self.branch2(x)
 74 |         x3 = self.branch3(x)
 75 |         out = torch.cat((x0, x1, x2, x3), 1)
 76 |         return out
 77 | 
 78 | 
 79 | class Mixed_4b(nn.Module):
 80 |     def __init__(self):
 81 |         super(Mixed_4b, self).__init__()
 82 | 
 83 |         self.branch0 = nn.Sequential(
 84 |             BasicConv3d(480, 192, kernel_size=1, stride=1),
 85 |         )
 86 |         self.branch1 = nn.Sequential(
 87 |             BasicConv3d(480, 96, kernel_size=1, stride=1),
 88 |             BasicConv3d(96, 208, kernel_size=3, stride=1, padding=1),
 89 |         )
 90 |         self.branch2 = nn.Sequential(
 91 |             BasicConv3d(480, 16, kernel_size=1, stride=1),
 92 |             BasicConv3d(16, 48, kernel_size=3, stride=1, padding=1),
 93 |         )
 94 |         self.branch3 = nn.Sequential(
 95 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
 96 |             BasicConv3d(480, 64, kernel_size=1, stride=1),
 97 |         )
 98 | 
 99 |     def forward(self, x):
100 |         x0 = self.branch0(x)
101 |         x1 = self.branch1(x)
102 |         x2 = self.branch2(x)
103 |         x3 = self.branch3(x)
104 |         out = torch.cat((x0, x1, x2, x3), 1)
105 |         return out
106 | 
107 | 
108 | class Mixed_4c(nn.Module):
109 |     def __init__(self):
110 |         super(Mixed_4c, self).__init__()
111 | 
112 |         self.branch0 = nn.Sequential(
113 |             BasicConv3d(512, 160, kernel_size=1, stride=1),
114 |         )
115 |         self.branch1 = nn.Sequential(
116 |             BasicConv3d(512, 112, kernel_size=1, stride=1),
117 |             BasicConv3d(112, 224, kernel_size=3, stride=1, padding=1),
118 |         )
119 |         self.branch2 = nn.Sequential(
120 |             BasicConv3d(512, 24, kernel_size=1, stride=1),
121 |             BasicConv3d(24, 64, kernel_size=3, stride=1, padding=1),
122 |         )
123 |         self.branch3 = nn.Sequential(
124 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
125 |             BasicConv3d(512, 64, kernel_size=1, stride=1),
126 |         )
127 | 
128 |     def forward(self, x):
129 |         x0 = self.branch0(x)
130 |         x1 = self.branch1(x)
131 |         x2 = self.branch2(x)
132 |         x3 = self.branch3(x)
133 |         out = torch.cat((x0, x1, x2, x3), 1)
134 |         return out
135 | 
136 | 
137 | class Mixed_4d(nn.Module):
138 |     def __init__(self):
139 |         super(Mixed_4d, self).__init__()
140 | 
141 |         self.branch0 = nn.Sequential(
142 |             BasicConv3d(512, 128, kernel_size=1, stride=1),
143 |         )
144 |         self.branch1 = nn.Sequential(
145 |             BasicConv3d(512, 128, kernel_size=1, stride=1),
146 |             BasicConv3d(128, 256, kernel_size=3, stride=1, padding=1),
147 |         )
148 |         self.branch2 = nn.Sequential(
149 |             BasicConv3d(512, 24, kernel_size=1, stride=1),
150 |             BasicConv3d(24, 64, kernel_size=3, stride=1, padding=1),
151 |         )
152 |         self.branch3 = nn.Sequential(
153 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
154 |             BasicConv3d(512, 64, kernel_size=1, stride=1),
155 |         )
156 | 
157 |     def forward(self, x):
158 |         x0 = self.branch0(x)
159 |         x1 = self.branch1(x)
160 |         x2 = self.branch2(x)
161 |         x3 = self.branch3(x)
162 |         out = torch.cat((x0, x1, x2, x3), 1)
163 |         return out
164 | 
165 | 
166 | class Mixed_4e(nn.Module):
167 |     def __init__(self):
168 |         super(Mixed_4e, self).__init__()
169 | 
170 |         self.branch0 = nn.Sequential(
171 |             BasicConv3d(512, 112, kernel_size=1, stride=1),
172 |         )
173 |         self.branch1 = nn.Sequential(
174 |             BasicConv3d(512, 144, kernel_size=1, stride=1),
175 |             BasicConv3d(144, 288, kernel_size=3, stride=1, padding=1),
176 |         )
177 |         self.branch2 = nn.Sequential(
178 |             BasicConv3d(512, 32, kernel_size=1, stride=1),
179 |             BasicConv3d(32, 64, kernel_size=3, stride=1, padding=1),
180 |         )
181 |         self.branch3 = nn.Sequential(
182 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
183 |             BasicConv3d(512, 64, kernel_size=1, stride=1),
184 |         )
185 | 
186 |     def forward(self, x):
187 |         x0 = self.branch0(x)
188 |         x1 = self.branch1(x)
189 |         x2 = self.branch2(x)
190 |         x3 = self.branch3(x)
191 |         out = torch.cat((x0, x1, x2, x3), 1)
192 |         return out
193 | 
194 | 
195 | class Mixed_4f(nn.Module):
196 |     def __init__(self):
197 |         super(Mixed_4f, self).__init__()
198 | 
199 |         self.branch0 = nn.Sequential(
200 |             BasicConv3d(528, 256, kernel_size=1, stride=1),
201 |         )
202 |         self.branch1 = nn.Sequential(
203 |             BasicConv3d(528, 160, kernel_size=1, stride=1),
204 |             BasicConv3d(160, 320, kernel_size=3, stride=1, padding=1),
205 |         )
206 |         self.branch2 = nn.Sequential(
207 |             BasicConv3d(528, 32, kernel_size=1, stride=1),
208 |             BasicConv3d(32, 128, kernel_size=3, stride=1, padding=1),
209 |         )
210 |         self.branch3 = nn.Sequential(
211 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
212 |             BasicConv3d(528, 128, kernel_size=1, stride=1),
213 |         )
214 | 
215 |     def forward(self, x):
216 |         x0 = self.branch0(x)
217 |         x1 = self.branch1(x)
218 |         x2 = self.branch2(x)
219 |         x3 = self.branch3(x)
220 |         out = torch.cat((x0, x1, x2, x3), 1)
221 |         return out
222 | 
223 | 
224 | class Mixed_5b(nn.Module):
225 |     def __init__(self):
226 |         super(Mixed_5b, self).__init__()
227 | 
228 |         self.branch0 = nn.Sequential(
229 |             BasicConv3d(832, 256, kernel_size=1, stride=1),
230 |         )
231 |         self.branch1 = nn.Sequential(
232 |             BasicConv3d(832, 160, kernel_size=1, stride=1),
233 |             BasicConv3d(160, 320, kernel_size=3, stride=1, padding=1),
234 |         )
235 |         self.branch2 = nn.Sequential(
236 |             BasicConv3d(832, 32, kernel_size=1, stride=1),
237 |             BasicConv3d(32, 128, kernel_size=3, stride=1, padding=1),
238 |         )
239 |         self.branch3 = nn.Sequential(
240 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
241 |             BasicConv3d(832, 128, kernel_size=1, stride=1),
242 |         )
243 | 
244 |     def forward(self, x):
245 |         x0 = self.branch0(x)
246 |         x1 = self.branch1(x)
247 |         x2 = self.branch2(x)
248 |         x3 = self.branch3(x)
249 |         out = torch.cat((x0, x1, x2, x3), 1)
250 |         return out
251 | 
252 | 
253 | class Mixed_5c(nn.Module):
254 |     def __init__(self):
255 |         super(Mixed_5c, self).__init__()
256 | 
257 |         self.branch0 = nn.Sequential(
258 |             BasicConv3d(832, 384, kernel_size=1, stride=1),
259 |         )
260 |         self.branch1 = nn.Sequential(
261 |             BasicConv3d(832, 192, kernel_size=1, stride=1),
262 |             BasicConv3d(192, 384, kernel_size=3, stride=1, padding=1),
263 |         )
264 |         self.branch2 = nn.Sequential(
265 |             BasicConv3d(832, 48, kernel_size=1, stride=1),
266 |             BasicConv3d(48, 128, kernel_size=3, stride=1, padding=1),
267 |         )
268 |         self.branch3 = nn.Sequential(
269 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=1, padding=1),
270 |             BasicConv3d(832, 128, kernel_size=1, stride=1),
271 |         )
272 | 
273 |     def forward(self, x):
274 |         x0 = self.branch0(x)
275 |         x1 = self.branch1(x)
276 |         x2 = self.branch2(x)
277 |         x3 = self.branch3(x)
278 |         out = torch.cat((x0, x1, x2, x3), 1)
279 |         return out
280 | 
281 | class I3D(nn.Module):
282 | 
283 |     def __init__(self, num_classes=400, dropout_keep_prob = 1, input_channel = 3, spatial_squeeze=True):
284 |         super(I3D, self).__init__()
285 |         self.features = nn.Sequential(
286 |             BasicConv3d(input_channel, 64, kernel_size=7, stride=2, padding=3), # (64, 32, 112, 112)
287 |             nn.MaxPool3d(kernel_size=(1,3,3), stride=(1,2,2), padding=(0,1,1)),  # (64, 32, 56, 56)
288 |             BasicConv3d(64, 64, kernel_size=1, stride=1), # (64, 32, 56, 56)
289 |             BasicConv3d(64, 192, kernel_size=3, stride=1, padding=1),  # (192, 32, 56, 56)
290 |             nn.MaxPool3d(kernel_size=(1,3,3), stride=(1,2,2), padding=(0,1,1)),  # (192, 32, 28, 28)
291 |             Mixed_3b(), # (256, 32, 28, 28)
292 |             Mixed_3c(), # (480, 32, 28, 28)
293 |             nn.MaxPool3d(kernel_size=(3, 3, 3), stride=(2, 2, 2), padding=(1, 1, 1)), # (480, 16, 14, 14)
294 |             Mixed_4b(),# (512, 16, 14, 14)
295 |             Mixed_4c(),# (512, 16, 14, 14)
296 |             Mixed_4d(),# (512, 16, 14, 14)
297 |             Mixed_4e(),# (528, 16, 14, 14)
298 |             Mixed_4f(),# (832, 16, 14, 14)
299 |             nn.MaxPool3d(kernel_size=(2, 2, 2), stride=(2, 2, 2), padding=(0, 0, 0)), # (832, 8, 7, 7)
300 |             Mixed_5b(), # (832, 8, 7, 7)
301 |             Mixed_5c(), # (1024, 8, 7, 7)
302 |             nn.AvgPool3d(kernel_size=(2, 7, 7), stride=1),# (1024, 8, 1, 1)
303 |             nn.Dropout3d(dropout_keep_prob),
304 |             nn.Conv3d(1024, num_classes, kernel_size=1, stride=1, bias=True),# (400, 8, 1, 1)
305 |         )
306 |         self.spatial_squeeze = spatial_squeeze
307 |         self.softmax = nn.Softmax()
308 | 
309 |     def forward(self, x):
310 |         logits = self.features(x)
311 | 
312 |         if self.spatial_squeeze:
313 |             logits = logits.squeeze(3)
314 |             logits = logits.squeeze(3)
315 | 
316 |         averaged_logits = torch.mean(logits, 2)
317 |         predictions = self.softmax(averaged_logits)
318 | 
319 |         return predictions, averaged_logits
320 | 
321 | if __name__ == '__main__':
322 | 
323 |     model = I3D()
324 |     model.cuda()
325 |     model = model.eval()
326 |     data=torch.autograd.Variable(torch.rand(10,3,64,224,224)).cuda()   # if modality=='Flow', please change the 2nd dimension 3==>2
327 |     out=model(data)
328 |     # print (model)
329 |     print (out[0].size())


--------------------------------------------------------------------------------
/models/i3dpt.py:
--------------------------------------------------------------------------------
  1 | import math
  2 | import os
  3 | 
  4 | import numpy as np
  5 | import torch
  6 | import torch.nn
  7 | from torch.nn import ReplicationPad3d
  8 | from torch.nn.modules.padding import ConstantPad3d
  9 | 
 10 | def get_padding_shape(filter_shape, stride):
 11 |     def _pad_top_bottom(filter_dim, stride_val):
 12 |         pad_along = max(filter_dim - stride_val, 0)
 13 |         pad_top = pad_along // 2
 14 |         pad_bottom = pad_along - pad_top
 15 |         return pad_top, pad_bottom
 16 | 
 17 |     padding_shape = []
 18 |     for filter_dim, stride_val in zip(filter_shape, stride):
 19 |         pad_top, pad_bottom = _pad_top_bottom(filter_dim, stride_val)
 20 |         padding_shape.append(pad_top)
 21 |         padding_shape.append(pad_bottom)
 22 |     depth_top = padding_shape.pop(0)
 23 |     depth_bottom = padding_shape.pop(0)
 24 |     padding_shape.append(depth_top)
 25 |     padding_shape.append(depth_bottom)
 26 | 
 27 |     return tuple(padding_shape)
 28 | 
 29 | 
 30 | def simplify_padding(padding_shapes):
 31 |     all_same = True
 32 |     padding_init = padding_shapes[0]
 33 |     for pad in padding_shapes[1:]:
 34 |         if pad != padding_init:
 35 |             all_same = False
 36 |     return all_same, padding_init
 37 | 
 38 | 
 39 | class Unit3Dpy(torch.nn.Module):
 40 |     def __init__(self,
 41 |                  in_channels,
 42 |                  out_channels,
 43 |                  kernel_size=(1, 1, 1),
 44 |                  stride=(1, 1, 1),
 45 |                  activation='relu',
 46 |                  padding='SAME',
 47 |                  use_bias=False,
 48 |                  use_bn=True):
 49 |         super(Unit3Dpy, self).__init__()
 50 | 
 51 |         self.padding = padding
 52 |         self.activation = activation
 53 |         self.use_bn = use_bn
 54 |         if padding == 'SAME':
 55 |             padding_shape = get_padding_shape(kernel_size, stride)
 56 |             simplify_pad, pad_size = simplify_padding(padding_shape)
 57 |             self.simplify_pad = simplify_pad
 58 |         elif padding == 'VALID':
 59 |             padding_shape = 0
 60 |         else:
 61 |             raise ValueError(
 62 |                 'padding should be in [VALID|SAME] but got {}'.format(padding))
 63 | 
 64 |         if padding == 'SAME':
 65 |             if not simplify_pad:
 66 |                 self.pad = torch.nn.ConstantPad3d(padding_shape, 0)
 67 |                 self.conv3d = torch.nn.Conv3d(
 68 |                     in_channels,
 69 |                     out_channels,
 70 |                     kernel_size,
 71 |                     stride=stride,
 72 |                     bias=use_bias)
 73 |             else:
 74 |                 self.conv3d = torch.nn.Conv3d(
 75 |                     in_channels,
 76 |                     out_channels,
 77 |                     kernel_size,
 78 |                     stride=stride,
 79 |                     padding=pad_size,
 80 |                     bias=use_bias)
 81 |         elif padding == 'VALID':
 82 |             self.conv3d = torch.nn.Conv3d(
 83 |                 in_channels,
 84 |                 out_channels,
 85 |                 kernel_size,
 86 |                 padding=padding_shape,
 87 |                 stride=stride,
 88 |                 bias=use_bias)
 89 |         else:
 90 |             raise ValueError(
 91 |                 'padding should be in [VALID|SAME] but got {}'.format(padding))
 92 | 
 93 |         if self.use_bn:
 94 |             self.batch3d = torch.nn.BatchNorm3d(out_channels)
 95 | 
 96 |         if activation == 'relu':
 97 |             self.activation = torch.nn.functional.relu
 98 | 
 99 |     def forward(self, inp):
100 |         if self.padding == 'SAME' and self.simplify_pad is False:
101 |             inp = self.pad(inp)
102 |         out = self.conv3d(inp)
103 |         if self.use_bn:
104 |             out = self.batch3d(out)
105 |         if self.activation is not None:
106 |             out = torch.nn.functional.relu(out)
107 |         return out
108 | 
109 | 
110 | class MaxPool3dTFPadding(torch.nn.Module):
111 |     def __init__(self, kernel_size, stride=None, padding='SAME'):
112 |         super(MaxPool3dTFPadding, self).__init__()
113 |         if padding == 'SAME':
114 |             padding_shape = get_padding_shape(kernel_size, stride)
115 |             self.padding_shape = padding_shape
116 |             self.pad = torch.nn.ConstantPad3d(padding_shape, 0)
117 |         self.pool = torch.nn.MaxPool3d(kernel_size, stride)
118 | 
119 |     def forward(self, inp):
120 |         inp = self.pad(inp)
121 |         out = self.pool(inp)
122 |         return out
123 | 
124 | 
125 | class Mixed(torch.nn.Module):
126 |     def __init__(self, in_channels, out_channels):
127 |         super(Mixed, self).__init__()
128 |         # Branch 0
129 |         self.branch_0 = Unit3Dpy(
130 |             in_channels, out_channels[0], kernel_size=(1, 1, 1))
131 | 
132 |         # Branch 1
133 |         branch_1_conv1 = Unit3Dpy(
134 |             in_channels, out_channels[1], kernel_size=(1, 1, 1))
135 |         branch_1_conv2 = Unit3Dpy(
136 |             out_channels[1], out_channels[2], kernel_size=(3, 3, 3))
137 |         self.branch_1 = torch.nn.Sequential(branch_1_conv1, branch_1_conv2)
138 | 
139 |         # Branch 2
140 |         branch_2_conv1 = Unit3Dpy(
141 |             in_channels, out_channels[3], kernel_size=(1, 1, 1))
142 |         branch_2_conv2 = Unit3Dpy(
143 |             out_channels[3], out_channels[4], kernel_size=(3, 3, 3))
144 |         self.branch_2 = torch.nn.Sequential(branch_2_conv1, branch_2_conv2)
145 | 
146 |         # Branch3
147 |         branch_3_pool = MaxPool3dTFPadding(
148 |             kernel_size=(3, 3, 3), stride=(1, 1, 1), padding='SAME')
149 |         branch_3_conv2 = Unit3Dpy(
150 |             in_channels, out_channels[5], kernel_size=(1, 1, 1))
151 |         self.branch_3 = torch.nn.Sequential(branch_3_pool, branch_3_conv2)
152 | 
153 |     def forward(self, inp):
154 |         out_0 = self.branch_0(inp)
155 |         out_1 = self.branch_1(inp)
156 |         out_2 = self.branch_2(inp)
157 |         out_3 = self.branch_3(inp)
158 |         out = torch.cat((out_0, out_1, out_2, out_3), 1)
159 |         return out
160 | 
161 | 
162 | class I3D(torch.nn.Module):
163 |     def __init__(self,
164 |                  num_classes,
165 |                  modality='rgb',
166 |                  dropout_keep_prob=1,
167 |                  name='inception'):
168 |         super(I3D, self).__init__()
169 | 
170 |         self.name = name
171 |         self.num_classes = num_classes
172 |         if modality == 'rgb':
173 |             in_channels = 3
174 |         elif modality == 'flow':
175 |             in_channels = 2
176 |         else:
177 |             raise ValueError(
178 |                 '{} not among known modalities [rgb|flow]'.format(modality))
179 |         self.modality = modality
180 | 
181 |         conv3d_1a_7x7 = Unit3Dpy(
182 |             out_channels=64,
183 |             in_channels=in_channels,
184 |             kernel_size=(7, 7, 7),
185 |             stride=(2, 2, 2),
186 |             padding='SAME')
187 |         # 1st conv-pool
188 |         self.conv3d_1a_7x7 = conv3d_1a_7x7
189 |         self.maxPool3d_2a_3x3 = MaxPool3dTFPadding(
190 |             kernel_size=(1, 3, 3), stride=(1, 2, 2), padding='SAME')
191 |         # conv conv
192 |         conv3d_2b_1x1 = Unit3Dpy(
193 |             out_channels=64,
194 |             in_channels=64,
195 |             kernel_size=(1, 1, 1),
196 |             padding='SAME')
197 |         self.conv3d_2b_1x1 = conv3d_2b_1x1
198 |         conv3d_2c_3x3 = Unit3Dpy(
199 |             out_channels=192,
200 |             in_channels=64,
201 |             kernel_size=(3, 3, 3),
202 |             padding='SAME')
203 |         self.conv3d_2c_3x3 = conv3d_2c_3x3
204 |         self.maxPool3d_3a_3x3 = MaxPool3dTFPadding(
205 |             kernel_size=(1, 3, 3), stride=(1, 2, 2), padding='SAME')
206 | 
207 |         # Mixed_3b
208 |         self.mixed_3b = Mixed(192, [64, 96, 128, 16, 32, 32])
209 |         self.mixed_3c = Mixed(256, [128, 128, 192, 32, 96, 64])
210 | 
211 |         self.maxPool3d_4a_3x3 = MaxPool3dTFPadding(
212 |             kernel_size=(3, 3, 3), stride=(2, 2, 2), padding='SAME')
213 | 
214 |         # Mixed 4
215 |         self.mixed_4b = Mixed(480, [192, 96, 208, 16, 48, 64])
216 |         self.mixed_4c = Mixed(512, [160, 112, 224, 24, 64, 64])
217 |         self.mixed_4d = Mixed(512, [128, 128, 256, 24, 64, 64])
218 |         self.mixed_4e = Mixed(512, [112, 144, 288, 32, 64, 64])
219 |         self.mixed_4f = Mixed(528, [256, 160, 320, 32, 128, 128])
220 | 
221 |         # Ugly hack because I didn't use tensorflow's exact padding function
222 |         self.pad_5a = torch.nn.ConstantPad3d((0, 0, 0, 0, 0, 1), 0)
223 |         self.maxPool3d_5a_2x2 = MaxPool3dTFPadding(
224 |             kernel_size=(2, 2, 2), stride=(2, 2, 2), padding='SAME')
225 | 
226 |         # Mixed 5
227 |         self.mixed_5b = Mixed(832, [256, 160, 320, 32, 128, 128])
228 |         self.mixed_5c = Mixed(832, [384, 192, 384, 48, 128, 128])
229 | 
230 |         self.avg_pool = torch.nn.AvgPool3d((2, 7, 7), (1, 1, 1))
231 |         self.dropout = torch.nn.Dropout(dropout_keep_prob)
232 |         self.conv3d_0c_1x1 = Unit3Dpy(
233 |             in_channels=1024,
234 |             out_channels=self.num_classes,
235 |             kernel_size=(1, 1, 1),
236 |             activation=None,
237 |             use_bias=True,
238 |             use_bn=False)
239 |         self.softmax = torch.nn.Softmax(1)
240 | 
241 |     def forward(self, inp):
242 |         # Preprocessing
243 |         out = self.conv3d_1a_7x7(inp)
244 |         out = self.maxPool3d_2a_3x3(out)
245 |         out = self.conv3d_2b_1x1(out)
246 |         out = self.conv3d_2c_3x3(out)
247 |         out = self.maxPool3d_3a_3x3(out)
248 |         out = self.mixed_3b(out)
249 |         out = self.mixed_3c(out)
250 |         out = self.maxPool3d_4a_3x3(out)
251 |         out = self.mixed_4b(out)
252 |         out = self.mixed_4c(out)
253 |         out = self.mixed_4d(out)
254 |         out = self.mixed_4e(out)
255 |         out = self.mixed_4f(out)
256 |         out = self.pad_5a(out)
257 |         out = self.maxPool3d_5a_2x2(out)
258 |         out = self.mixed_5b(out)
259 |         out = self.mixed_5c(out)
260 |         out = self.avg_pool(out)
261 |         out = self.dropout(out)
262 |         out = self.conv3d_0c_1x1(out)
263 |         out = out.squeeze(3)
264 |         out = out.squeeze(3)
265 |         out = out.mean(2)
266 |         out_logits = out
267 |         out = self.softmax(out_logits)
268 |         return out, out_logits
269 | 
270 |     def load_tf_weights(self, sess):
271 |         state_dict = {}
272 |         if self.modality == 'rgb':
273 |             prefix = 'RGB/inception_i3d'
274 |         elif self.modality == 'flow':
275 |             prefix = 'Flow/inception_i3d'
276 |         load_conv3d(state_dict, 'conv3d_1a_7x7', sess,
277 |                     os.path.join(prefix, 'Conv3d_1a_7x7'))
278 |         load_conv3d(state_dict, 'conv3d_2b_1x1', sess,
279 |                     os.path.join(prefix, 'Conv3d_2b_1x1'))
280 |         load_conv3d(state_dict, 'conv3d_2c_3x3', sess,
281 |                     os.path.join(prefix, 'Conv3d_2c_3x3'))
282 | 
283 |         load_mixed(state_dict, 'mixed_3b', sess,
284 |                    os.path.join(prefix, 'Mixed_3b'))
285 |         load_mixed(state_dict, 'mixed_3c', sess,
286 |                    os.path.join(prefix, 'Mixed_3c'))
287 |         load_mixed(state_dict, 'mixed_4b', sess,
288 |                    os.path.join(prefix, 'Mixed_4b'))
289 |         load_mixed(state_dict, 'mixed_4c', sess,
290 |                    os.path.join(prefix, 'Mixed_4c'))
291 |         load_mixed(state_dict, 'mixed_4d', sess,
292 |                    os.path.join(prefix, 'Mixed_4d'))
293 |         load_mixed(state_dict, 'mixed_4e', sess,
294 |                    os.path.join(prefix, 'Mixed_4e'))
295 |         # Here goest to 0.1 max error with tf
296 |         load_mixed(state_dict, 'mixed_4f', sess,
297 |                    os.path.join(prefix, 'Mixed_4f'))
298 | 
299 |         load_mixed(
300 |             state_dict,
301 |             'mixed_5b',
302 |             sess,
303 |             os.path.join(prefix, 'Mixed_5b'),
304 |             fix_typo=True)
305 |         load_mixed(state_dict, 'mixed_5c', sess,
306 |                    os.path.join(prefix, 'Mixed_5c'))
307 |         load_conv3d(
308 |             state_dict,
309 |             'conv3d_0c_1x1',
310 |             sess,
311 |             os.path.join(prefix, 'Logits', 'Conv3d_0c_1x1'),
312 |             bias=True,
313 |             bn=False)
314 |         self.load_state_dict(state_dict)
315 | 
316 | 
317 | def get_conv_params(sess, name, bias=False):
318 |     # Get conv weights
319 |     conv_weights_tensor = sess.graph.get_tensor_by_name(
320 |         os.path.join(name, 'w:0'))
321 |     if bias:
322 |         conv_bias_tensor = sess.graph.get_tensor_by_name(
323 |             os.path.join(name, 'b:0'))
324 |         conv_bias = sess.run(conv_bias_tensor)
325 |     conv_weights = sess.run(conv_weights_tensor)
326 |     conv_shape = conv_weights.shape
327 | 
328 |     kernel_shape = conv_shape[0:3]
329 |     in_channels = conv_shape[3]
330 |     out_channels = conv_shape[4]
331 | 
332 |     conv_op = sess.graph.get_operation_by_name(
333 |         os.path.join(name, 'convolution'))
334 |     padding_name = conv_op.get_attr('padding')
335 |     padding = _get_padding(padding_name, kernel_shape)
336 |     all_strides = conv_op.get_attr('strides')
337 |     strides = all_strides[1:4]
338 |     conv_params = [
339 |         conv_weights, kernel_shape, in_channels, out_channels, strides, padding
340 |     ]
341 |     if bias:
342 |         conv_params.append(conv_bias)
343 |     return conv_params
344 | 
345 | 
346 | def get_bn_params(sess, name):
347 |     moving_mean_tensor = sess.graph.get_tensor_by_name(
348 |         os.path.join(name, 'moving_mean:0'))
349 |     moving_var_tensor = sess.graph.get_tensor_by_name(
350 |         os.path.join(name, 'moving_variance:0'))
351 |     beta_tensor = sess.graph.get_tensor_by_name(os.path.join(name, 'beta:0'))
352 |     moving_mean = sess.run(moving_mean_tensor)
353 |     moving_var = sess.run(moving_var_tensor)
354 |     beta = sess.run(beta_tensor)
355 |     return moving_mean, moving_var, beta
356 | 
357 | 
358 | def _get_padding(padding_name, conv_shape):
359 |     padding_name = padding_name.decode("utf-8")
360 |     if padding_name == "VALID":
361 |         return [0, 0]
362 |     elif padding_name == "SAME":
363 |         #return [math.ceil(int(conv_shape[0])/2), math.ceil(int(conv_shape[1])/2)]
364 |         return [
365 |             math.floor(int(conv_shape[0]) / 2),
366 |             math.floor(int(conv_shape[1]) / 2),
367 |             math.floor(int(conv_shape[2]) / 2)
368 |         ]
369 |     else:
370 |         raise ValueError('Invalid padding name ' + padding_name)
371 | 
372 | 
373 | def load_conv3d(state_dict, name_pt, sess, name_tf, bias=False, bn=True):
374 |     # Transfer convolution params
375 |     conv_name_tf = os.path.join(name_tf, 'conv_3d')
376 |     conv_params = get_conv_params(sess, conv_name_tf, bias=bias)
377 |     if bias:
378 |         conv_weights, kernel_shape, in_channels, out_channels, strides, padding, conv_bias = conv_params
379 |     else:
380 |         conv_weights, kernel_shape, in_channels, out_channels, strides, padding = conv_params
381 | 
382 |     conv_weights_rs = np.transpose(
383 |         conv_weights, (4, 3, 0, 1,
384 |                        2))  # to pt format (out_c, in_c, depth, height, width)
385 |     state_dict[name_pt + '.conv3d.weight'] = torch.from_numpy(conv_weights_rs)
386 |     if bias:
387 |         state_dict[name_pt + '.conv3d.bias'] = torch.from_numpy(conv_bias)
388 | 
389 |     # Transfer batch norm params
390 |     if bn:
391 |         conv_tf_name = os.path.join(name_tf, 'batch_norm')
392 |         moving_mean, moving_var, beta = get_bn_params(sess, conv_tf_name)
393 | 
394 |         out_planes = conv_weights_rs.shape[0]
395 |         state_dict[name_pt + '.batch3d.weight'] = torch.ones(out_planes)
396 |         state_dict[name_pt + '.batch3d.bias'] = torch.from_numpy(beta)
397 |         state_dict[name_pt
398 |                    + '.batch3d.running_mean'] = torch.from_numpy(moving_mean)
399 |         state_dict[name_pt
400 |                    + '.batch3d.running_var'] = torch.from_numpy(moving_var)
401 | 
402 | 
403 | def load_mixed(state_dict, name_pt, sess, name_tf, fix_typo=False):
404 |     # Branch 0
405 |     load_conv3d(state_dict, name_pt + '.branch_0', sess,
406 |                 os.path.join(name_tf, 'Branch_0/Conv3d_0a_1x1'))
407 | 
408 |     # Branch .1
409 |     load_conv3d(state_dict, name_pt + '.branch_1.0', sess,
410 |                 os.path.join(name_tf, 'Branch_1/Conv3d_0a_1x1'))
411 |     load_conv3d(state_dict, name_pt + '.branch_1.1', sess,
412 |                 os.path.join(name_tf, 'Branch_1/Conv3d_0b_3x3'))
413 | 
414 |     # Branch 2
415 |     load_conv3d(state_dict, name_pt + '.branch_2.0', sess,
416 |                 os.path.join(name_tf, 'Branch_2/Conv3d_0a_1x1'))
417 |     if fix_typo:
418 |         load_conv3d(state_dict, name_pt + '.branch_2.1', sess,
419 |                     os.path.join(name_tf, 'Branch_2/Conv3d_0a_3x3'))
420 |     else:
421 |         load_conv3d(state_dict, name_pt + '.branch_2.1', sess,
422 |                     os.path.join(name_tf, 'Branch_2/Conv3d_0b_3x3'))
423 | 
424 |     # Branch 3
425 |     load_conv3d(state_dict, name_pt + '.branch_3.1', sess,
426 |                 os.path.join(name_tf, 'Branch_3/Conv3d_0b_1x1'))
427 | 
428 | if __name__ == '__main__':
429 | 
430 | 	model = I3D(num_classes=400, modality='rgb')
431 | 	data=torch.autograd.Variable(torch.rand(10,3,16,224,224))   # if modality=='Flow', please change the 2nd dimension 3==>2
432 | 	out=model(data)
433 | 	print (model)
434 | 	print (out[0].size())


--------------------------------------------------------------------------------
/models/p3d_model.py:
--------------------------------------------------------------------------------
  1 | from __future__ import print_function
  2 | import torch
  3 | import torch.nn as nn
  4 | import numpy as np
  5 | import torch.nn.functional as F
  6 | from torch.autograd import Variable
  7 | import math
  8 | from functools import partial
  9 | import logging
 10 | 
 11 | __all__ = ['P3D', 'P3D63', 'P3D131', 'P3D199']
 12 | 
 13 | 
 14 | def conv_S(in_planes, out_planes, stride=1, padding=1):
 15 |     # as is descriped, conv S is 1x3x3
 16 |     return nn.Conv3d(in_planes, out_planes, kernel_size=(1, 3, 3), stride=1,
 17 |                      padding=padding, bias=False)
 18 | 
 19 | 
 20 | def conv_T(in_planes, out_planes, stride=1, padding=1):
 21 |     # conv T is 3x1x1
 22 |     return nn.Conv3d(in_planes, out_planes, kernel_size=(3, 1, 1), stride=1,
 23 |                      padding=padding, bias=False)
 24 | 
 25 | 
 26 | def downsample_basic_block(x, planes, stride):
 27 |     out = F.avg_pool3d(x, kernel_size=1, stride=stride)
 28 |     zero_pads = torch.Tensor(out.size(0), planes - out.size(1),
 29 |                              out.size(2), out.size(3),
 30 |                              out.size(4)).zero_()
 31 |     if isinstance(out.data, torch.cuda.FloatTensor):
 32 |         zero_pads = zero_pads.cuda()
 33 | 
 34 |     out = Variable(torch.cat([out.data, zero_pads], dim=1))
 35 | 
 36 |     return out
 37 | 
 38 | 
 39 | class Bottleneck(nn.Module):
 40 |     expansion = 4
 41 | 
 42 |     def __init__(self, inplanes, planes, stride=1, downsample=None, n_s=0, depth_3d=47, ST_struc=('A', 'B', 'C')):
 43 |         super(Bottleneck, self).__init__()
 44 |         self.downsample = downsample
 45 |         self.depth_3d = depth_3d
 46 |         self.ST_struc = ST_struc
 47 |         self.len_ST = len(self.ST_struc)
 48 | 
 49 |         stride_p = stride
 50 |         if not self.downsample == None:
 51 |             stride_p = (1, 2, 2)
 52 |         if n_s < self.depth_3d:
 53 |             if n_s == 0:
 54 |                 stride_p = 1
 55 |             self.conv1 = nn.Conv3d(inplanes, planes, kernel_size=1, bias=False, stride=stride_p)
 56 |             self.bn1 = nn.BatchNorm3d(planes)
 57 |         else:
 58 |             if n_s == self.depth_3d:
 59 |                 stride_p = 2
 60 |             else:
 61 |                 stride_p = 1
 62 |             self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False, stride=stride_p)
 63 |             self.bn1 = nn.BatchNorm2d(planes)
 64 |         # self.conv2 = nn.Conv3d(planes, planes, kernel_size=3, stride=stride,
 65 |         #                        padding=1, bias=False)
 66 |         self.id = n_s
 67 |         self.ST = list(self.ST_struc)[self.id % self.len_ST]
 68 |         if self.id < self.depth_3d:
 69 |             self.conv2 = conv_S(planes, planes, stride=1, padding=(0, 1, 1))
 70 |             self.bn2 = nn.BatchNorm3d(planes)
 71 |             #
 72 |             self.conv3 = conv_T(planes, planes, stride=1, padding=(1, 0, 0))
 73 |             self.bn3 = nn.BatchNorm3d(planes)
 74 |         else:
 75 |             self.conv_normal = nn.Conv2d(planes, planes, kernel_size=3, stride=1, padding=1, bias=False)
 76 |             self.bn_normal = nn.BatchNorm2d(planes)
 77 | 
 78 |         if n_s < self.depth_3d:
 79 |             self.conv4 = nn.Conv3d(planes, planes * 4, kernel_size=1, bias=False)
 80 |             self.bn4 = nn.BatchNorm3d(planes * 4)
 81 |         else:
 82 |             self.conv4 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
 83 |             self.bn4 = nn.BatchNorm2d(planes * 4)
 84 |         self.relu = nn.ReLU(inplace=True)
 85 | 
 86 |         self.stride = stride
 87 | 
 88 |     def ST_A(self, x):
 89 |         x = self.conv2(x)
 90 |         x = self.bn2(x)
 91 |         x = self.relu(x)
 92 | 
 93 |         x = self.conv3(x)
 94 |         x = self.bn3(x)
 95 |         x = self.relu(x)
 96 | 
 97 |         return x
 98 | 
 99 |     def ST_B(self, x):
100 |         tmp_x = self.conv2(x)
101 |         tmp_x = self.bn2(tmp_x)
102 |         tmp_x = self.relu(tmp_x)
103 | 
104 |         x = self.conv3(x)
105 |         x = self.bn3(x)
106 |         x = self.relu(x)
107 | 
108 |         return x + tmp_x
109 | 
110 |     def ST_C(self, x):
111 |         x = self.conv2(x)
112 |         x = self.bn2(x)
113 |         x = self.relu(x)
114 | 
115 |         tmp_x = self.conv3(x)
116 |         tmp_x = self.bn3(tmp_x)
117 |         tmp_x = self.relu(tmp_x)
118 | 
119 |         return x + tmp_x
120 | 
121 |     def forward(self, x):
122 |         residual = x
123 | 
124 |         out = self.conv1(x)
125 |         out = self.bn1(out)
126 |         out = self.relu(out)
127 | 
128 |         # out = self.conv2(out)
129 |         # out = self.bn2(out)
130 |         # out = self.relu(out)
131 |         if self.id < self.depth_3d:  # C3D parts:
132 | 
133 |             if self.ST == 'A':
134 |                 out = self.ST_A(out)
135 |             elif self.ST == 'B':
136 |                 out = self.ST_B(out)
137 |             elif self.ST == 'C':
138 |                 out = self.ST_C(out)
139 |         else:
140 |             out = self.conv_normal(out)  # normal is res5 part, C2D all.
141 |             out = self.bn_normal(out)
142 |             out = self.relu(out)
143 | 
144 |         out = self.conv4(out)
145 |         out = self.bn4(out)
146 | 
147 |         if self.downsample is not None:
148 |             residual = self.downsample(x)
149 | 
150 |         out += residual
151 |         out = self.relu(out)
152 | 
153 |         return out
154 | 
155 | 
156 | class P3D(nn.Module):
157 | 
158 |     def __init__(self, block, layers, modality='RGB',
159 |                  shortcut_type='B', num_classes=400, dropout=0.5, ST_struc=('A', 'B', 'C')):
160 | 
161 |         self.inplanes = 64
162 |         super(P3D, self).__init__()
163 |         # self.conv1 = nn.Conv3d(3, 64, kernel_size=7, stride=(1, 2, 2),
164 |         #                        padding=(3, 3, 3), bias=False)
165 |         self.input_channel = 3 if modality == 'RGB' else 2  # 2 is for flow
166 |         self.ST_struc = ST_struc
167 | 
168 |         self.conv1_custom = nn.Conv3d(self.input_channel, 64, kernel_size=(1, 7, 7), stride=(1, 2, 2),
169 |                                       padding=(0, 3, 3), bias=False)
170 | 
171 |         self.depth_3d = sum(layers[:3])  # C3D layers are only (res2,res3,res4),  res5 is C2D
172 | 
173 |         self.bn1 = nn.BatchNorm3d(64)  # bn1 is followed by conv1
174 |         self.cnt = 0
175 |         self.relu = nn.ReLU(inplace=True)
176 |         self.maxpool = nn.MaxPool3d(kernel_size=(2, 3, 3), stride=2, padding=(0, 1, 1))  # pooling layer for conv1.
177 |         self.maxpool_2 = nn.MaxPool3d(kernel_size=(2, 1, 1), padding=0,
178 |                                       stride=(2, 1, 1))  # pooling layer for res2, 3, 4.
179 | 
180 |         self.layer1 = self._make_layer(block, 64, layers[0], shortcut_type)
181 |         self.layer2 = self._make_layer(block, 128, layers[1], shortcut_type, stride=2)
182 |         self.layer3 = self._make_layer(block, 256, layers[2], shortcut_type, stride=2)
183 |         self.layer4 = self._make_layer(block, 512, layers[3], shortcut_type, stride=2)
184 | 
185 |         self.avgpool = nn.AvgPool2d(kernel_size=(5, 5), stride=1)  # pooling layer for res5.
186 |         self.dropout = nn.Dropout(p=dropout)
187 |         self.fc = nn.Linear(512 * block.expansion, num_classes)
188 | 
189 |         for m in self.modules():
190 |             if isinstance(m, nn.Conv3d):
191 |                 n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
192 |                 m.weight.data.normal_(0, math.sqrt(2. / n))
193 |             elif isinstance(m, nn.BatchNorm3d):
194 |                 m.weight.data.fill_(1)
195 |                 m.bias.data.zero_()
196 | 
197 |         # some private attribute
198 |         self.input_size = (self.input_channel, 16, 160, 160)  # input of the network
199 |         self.input_mean = [0.485, 0.456, 0.406] if modality == 'RGB' else [0.5]
200 |         self.input_std = [0.229, 0.224, 0.225] if modality == 'RGB' else [np.mean([0.229, 0.224, 0.225])]
201 | 
202 |     @property
203 |     def scale_size(self):
204 |         return self.input_size[2] * 256 // 160  # asume that raw images are resized (340,256).
205 | 
206 |     @property
207 |     def temporal_length(self):
208 |         return self.input_size[1]
209 | 
210 |     @property
211 |     def crop_size(self):
212 |         return self.input_size[2]
213 | 
214 |     def _make_layer(self, block, planes, blocks, shortcut_type, stride=1):
215 |         downsample = None
216 |         stride_p = stride  # especially for downsample branch.
217 | 
218 |         if self.cnt < self.depth_3d:
219 |             if self.cnt == 0:
220 |                 stride_p = 1
221 |             else:
222 |                 stride_p = (1, 2, 2)
223 |             if stride != 1 or self.inplanes != planes * block.expansion:
224 |                 if shortcut_type == 'A':
225 |                     downsample = partial(downsample_basic_block,
226 |                                          planes=planes * block.expansion,
227 |                                          stride=stride)
228 |                 else:
229 |                     downsample = nn.Sequential(
230 |                         nn.Conv3d(self.inplanes, planes * block.expansion,
231 |                                   kernel_size=1, stride=stride_p, bias=False),
232 |                         nn.BatchNorm3d(planes * block.expansion)
233 |                     )
234 | 
235 |         else:
236 |             if stride != 1 or self.inplanes != planes * block.expansion:
237 |                 if shortcut_type == 'A':
238 |                     downsample = partial(downsample_basic_block,
239 |                                          planes=planes * block.expansion,
240 |                                          stride=stride)
241 |                 else:
242 |                     downsample = nn.Sequential(
243 |                         nn.Conv2d(self.inplanes, planes * block.expansion,
244 |                                   kernel_size=1, stride=2, bias=False),
245 |                         nn.BatchNorm2d(planes * block.expansion)
246 |                     )
247 |         layers = []
248 |         layers.append(block(self.inplanes, planes, stride, downsample, n_s=self.cnt, depth_3d=self.depth_3d,
249 |                             ST_struc=self.ST_struc))
250 |         self.cnt += 1
251 | 
252 |         self.inplanes = planes * block.expansion
253 |         for i in range(1, blocks):
254 |             layers.append(block(self.inplanes, planes, n_s=self.cnt, depth_3d=self.depth_3d, ST_struc=self.ST_struc))
255 |             self.cnt += 1
256 | 
257 |         return nn.Sequential(*layers)
258 | 
259 |     def forward(self, x):
260 |         x = self.conv1_custom(x)
261 |         x = self.bn1(x)
262 |         x = self.relu(x)
263 |         x = self.maxpool(x)
264 | 
265 |         x = self.maxpool_2(self.layer1(x))  # Part Res2
266 |         x = self.maxpool_2(self.layer2(x))  # Part Res3
267 |         x = self.maxpool_2(self.layer3(x))  # Part Res4
268 | 
269 |         sizes = x.size()
270 |         x = x.view(-1, sizes[1], sizes[3], sizes[4])  # Part Res5
271 |         x = self.layer4(x)
272 |         x = self.avgpool(x)
273 | 
274 |         x = x.view(-1, self.fc.in_features)
275 |         x = self.fc(self.dropout(x))
276 | 
277 |         return x
278 | 
279 | 
280 | def P3D63(**kwargs):
281 |     """Construct a P3D63 modelbased on a ResNet-50-3D model.
282 |     """
283 |     model = P3D(Bottleneck, [3, 4, 6, 3], **kwargs)
284 |     return model
285 | 
286 | 
287 | def P3D131(**kwargs):
288 |     """Construct a P3D131 model based on a ResNet-101-3D model.
289 |     """
290 |     model = P3D(Bottleneck, [3, 4, 23, 3], **kwargs)
291 |     return model
292 | 
293 | 
294 | def P3D199(pretrained=False, modality='RGB', **kwargs):
295 |     """construct a P3D199 model based on a ResNet-152-3D model.
296 |     """
297 |     model = P3D(Bottleneck, [3, 8, 36, 3], modality=modality, **kwargs)
298 |     if pretrained == True:
299 |         if modality == 'RGB':
300 |             pretrained_file = 'p3d_rgb_199.checkpoint.pth.tar'
301 |         elif modality == 'Flow':
302 |             pretrained_file = 'p3d_flow_199.checkpoint.pth.tar'
303 |         weights = torch.load(pretrained_file)['state_dict']
304 |         print('loading trained model')
305 |         model.load_state_dict(weights)
306 |     return model
307 | 
308 | 
309 | # custom operation
310 | def get_optim_policies(model=None, modality='RGB', enable_pbn=True):
311 |     '''
312 |     first conv:         weight --> conv weight
313 |                         bias   --> conv bias
314 |     normal action:      weight --> non-first conv + fc weight
315 |                         bias   --> non-first conv + fc bias
316 |     bn:                 the first bn2, and many all bn3.
317 | 
318 |     '''
319 |     first_conv_weight = []
320 |     first_conv_bias = []
321 |     normal_weight = []
322 |     normal_bias = []
323 |     bn = []
324 | 
325 |     if model == None:
326 |         logging.info('no model!')
327 |         exit()
328 | 
329 |     conv_cnt = 0
330 |     bn_cnt = 0
331 |     for m in model.modules():
332 |         if isinstance(m, torch.nn.Conv3d) or isinstance(m, torch.nn.Conv2d):
333 |             ps = list(m.parameters())
334 |             conv_cnt += 1
335 |             if conv_cnt == 1:
336 |                 first_conv_weight.append(ps[0])
337 |                 if len(ps) == 2:
338 |                     first_conv_bias.append(ps[1])
339 |             else:
340 |                 normal_weight.append(ps[0])
341 |                 if len(ps) == 2:
342 |                     normal_bias.append(ps[1])
343 |         elif isinstance(m, torch.nn.Linear):
344 |             ps = list(m.parameters())
345 |             normal_weight.append(ps[0])
346 |             if len(ps) == 2:
347 |                 normal_bias.append(ps[1])
348 | 
349 |         elif isinstance(m, torch.nn.BatchNorm3d):
350 |             bn_cnt += 1
351 |             # later BN's are frozen
352 |             if not enable_pbn or bn_cnt == 1:
353 |                 bn.extend(list(m.parameters()))
354 |         elif isinstance(m, torch.nn.BatchNorm2d):
355 |             bn.extend(list(m.parameters()))
356 |         elif len(m._modules) == 0:
357 |             if len(list(m.parameters())) > 0:
358 |                 raise ValueError("New atomic module type: {}. Need to give it a learning policy".format(type(m)))
359 | 
360 |     slow_rate = 0.7
361 |     n_fore = int(len(normal_weight) * slow_rate)
362 |     slow_feat = normal_weight[:n_fore]  # finetune slowly.
363 |     slow_bias = normal_bias[:n_fore]
364 |     normal_feat = normal_weight[n_fore:]
365 |     normal_bias = normal_bias[n_fore:]
366 | 
367 |     return [
368 |         {'params': first_conv_weight, 'lr_mult': 5 if modality == 'Flow' else 1, 'decay_mult': 1,
369 |          'name': "first_conv_weight"},
370 |         {'params': first_conv_bias, 'lr_mult': 10 if modality == 'Flow' else 2, 'decay_mult': 0,
371 |          'name': "first_conv_bias"},
372 |         {'params': slow_feat, 'lr_mult': 1, 'decay_mult': 1,
373 |          'name': "slow_feat"},
374 |         {'params': slow_bias, 'lr_mult': 2, 'decay_mult': 0,
375 |          'name': "slow_bias"},
376 |         {'params': normal_feat, 'lr_mult': 1, 'decay_mult': 1,
377 |          'name': "normal_feat"},
378 |         {'params': normal_bias, 'lr_mult': 2, 'decay_mult': 0,
379 |          'name': "normal_bias"},
380 |         {'params': bn, 'lr_mult': 1, 'decay_mult': 0,
381 |          'name': "BN scale/shift"},
382 |     ]
383 | 
384 | 
385 | if __name__ == '__main__':
386 |     model = P3D199(pretrained=True, num_classes=400)
387 |     model = model.cuda()
388 |     data = torch.autograd.Variable(
389 |         torch.rand(10, 3, 16, 160, 160)).cuda()  # if modality=='Flow', please change the 2nd dimension 3==>2
390 |     out = model(data)
391 |     print(out.size(), out)


--------------------------------------------------------------------------------
/test.py:
--------------------------------------------------------------------------------
  1 | from __future__ import print_function
  2 | import os
  3 | import os.path
  4 | import time
  5 | import torch.backends.cudnn as cudnn
  6 | import torch.nn as nn
  7 | import torch.nn.parallel
  8 | import torch.optim as optim
  9 | import torch.utils.data as data
 10 | from meter import AverageMeter
 11 | 
 12 | from logger import Logger
 13 | # from video_transforms import *
 14 | from transforms import *
 15 | from Dataset import MyDataset
 16 | from models.p3d_model import P3D199, get_optim_policies
 17 | from models.C3D import C3D
 18 | from models.i3dpt import I3D
 19 | from utils import check_gpu, transfer_model, accuracy
 20 | 
 21 | class Testing(object):
 22 |     def __init__(self, name_list, num_classes=400, modality='RGB', **kwargs):
 23 |         self.__dict__.update(kwargs)
 24 | 
 25 |         self.num_classes = num_classes
 26 |         self.modality = modality
 27 |         self.name_list = name_list
 28 | 
 29 |         # Set best precision = 0
 30 |         self.best_prec1 = 0
 31 |         # init start epoch = 0
 32 |         self.start_epoch = 0
 33 | 
 34 |         self.checkDataFolder()
 35 | 
 36 |         self.loading_model()
 37 | 
 38 |         self.test_loader = self.loading_data()
 39 | 
 40 |         # run
 41 |         self.process()
 42 | 
 43 |     def checkDataFolder(self):
 44 |         try:
 45 |             os.stat('./' + self.model_type + '_' + self.data_set)
 46 |         except:
 47 |             os.mkdir('./' + self.model_type + '_' + self.data_set)
 48 |         self.data_folder = './' + self.model_type + '_' + self.data_set
 49 | 
 50 |     # Loading P3D model
 51 |     def loading_model(self):
 52 | 
 53 |         print('Loading %s model' % (self.model_type))
 54 |         if self.model_type == 'C3D':
 55 |             self.model = C3D()
 56 |         elif self.model_type == 'I3D':
 57 |             self.model = I3D(num_classes=400, modality='rgb')
 58 |         else:
 59 |             self.model = P3D199(pretrained=False, num_classes=400, dropout=self.dropout)
 60 | 
 61 | 
 62 |         # Transfer classes
 63 |         self.model = transfer_model(model=self.model, model_type=self.model_type, num_classes=self.num_classes)
 64 | 
 65 |         # Check gpu and run parallel
 66 |         if check_gpu() > 0:
 67 |             self.model = torch.nn.DataParallel(self.model).cuda()
 68 | 
 69 |         # define loss function (criterion) and optimizer
 70 |         if check_gpu() > 0:
 71 |             self.criterion = nn.CrossEntropyLoss().cuda()
 72 |         else:
 73 |             self.criterion = nn.CrossEntropyLoss()
 74 | 
 75 |         policies = get_optim_policies(model=self.model, modality=self.modality, enable_pbn=True)
 76 | 
 77 |         self.optimizer = optim.SGD(policies, lr=self.lr, momentum=self.momentum, weight_decay=self.weight_decay)
 78 | 
 79 |         file = os.path.join(self.data_folder, 'model_best.pth.tar')
 80 |         if os.path.isfile(file):
 81 |             print("=> loading checkpoint '{}'".format('model_best.pth.tar'))
 82 | 
 83 |             checkpoint = torch.load(file)
 84 |             self.start_epoch = checkpoint['epoch']
 85 |             self.best_prec1 = checkpoint['best_prec1']
 86 |             self.model.load_state_dict(checkpoint['state_dict'])
 87 |             self.optimizer.load_state_dict(checkpoint['optimizer'])
 88 |             print("=> loaded model best ")
 89 |         else:
 90 |             print("=> no model best found at ")
 91 |             exit()
 92 | 
 93 |         cudnn.benchmark = True
 94 | 
 95 |     # Loading data
 96 |     def loading_data(self):
 97 |         size = 160
 98 |         if self.model_type == 'C3D':
 99 |             size = 112
100 |         if self.model_type == 'I3D':
101 |             size = 224
102 |         normalize = Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
103 | 
104 |         val_transformations = Compose([
105 |             Resize((size, size)),
106 |             ToTensor(),
107 |             normalize
108 |         ])
109 | 
110 |         test_dataset = MyDataset(
111 |             self.data,
112 |             name_list=self.name_list,
113 |             data_folder="test",
114 |             version="1",
115 |             transform=val_transformations,
116 |             num_frames=self.num_frames
117 |         )
118 | 
119 |         test_loader = data.DataLoader(
120 |             test_dataset,
121 |             batch_size=self.batch_size,
122 |             shuffle=False,
123 |             num_workers=self.workers,
124 |             pin_memory=False)
125 | 
126 |         return test_loader
127 | 
128 |     # Test
129 |     def process(self):
130 |         acc = AverageMeter()
131 |         top1 = AverageMeter()
132 |         top5 = AverageMeter()
133 |         losses = AverageMeter()
134 |         log_file = os.path.join(self.data_folder, 'test.log')
135 |         logger = Logger('test', log_file)
136 |         # switch to evaluate mode
137 |         self.model.eval()
138 | 
139 |         start_time = time.clock()
140 |         print("Begin testing")
141 |         for i, (images, labels) in enumerate(self.test_loader):
142 |             if check_gpu() > 0:
143 |                 images = images.cuda(async=True)
144 |                 labels = labels.cuda(async=True)
145 | 
146 |             image_var = torch.autograd.Variable(images)
147 |             label_var = torch.autograd.Variable(labels)
148 | 
149 |             # compute y_pred
150 |             y_pred = self.model(image_var)
151 |             loss = self.criterion(y_pred, label_var)
152 | 
153 |             # measure accuracy and record loss
154 |             prec1, prec5 = accuracy(y_pred.data, labels, topk=(1, 5))
155 |             losses.update(loss.item(), images.size(0))
156 |             acc.update(prec1.item(), images.size(0))
157 |             top1.update(prec1.item(), images.size(0))
158 |             top5.update(prec5.item(), images.size(0))
159 | 
160 |             if i % self.print_freq == 0:
161 |                 print('TestVal: [{0}/{1}]\t'
162 |                       'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
163 |                       'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
164 |                       'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
165 |                     i, len(self.test_loader), loss=losses, top1=top1, top5=top5))
166 | 
167 |         print(
168 |             ' * Accuracy {acc.avg:.3f}  Acc@5 {top5.avg:.3f} Loss {loss.avg:.3f}'.format(acc=acc, top5=top5,
169 |                                                                                          loss=losses))
170 | 
171 |         end_time = time.clock()
172 |         print("Total testing time %.2gs" % (end_time - start_time))
173 |         logger.info("Total testing time %.2gs" % (end_time - start_time))
174 |         logger.info(
175 |             ' * Accuracy {acc.avg:.3f}  Acc@5 {top5.avg:.3f} Loss {loss.avg:.3f}'.format(acc=acc, top5=top5,
176 |                                                                                          loss=losses))
177 | 
178 | 
179 | 
180 | 
181 | 


--------------------------------------------------------------------------------
/train.py:
--------------------------------------------------------------------------------
  1 | from __future__ import print_function
  2 | import os
  3 | import os.path
  4 | import shutil
  5 | import time
  6 | import torch.backends.cudnn as cudnn
  7 | import torch.nn as nn
  8 | import torch.nn.parallel
  9 | import torch.optim as optim
 10 | import torch.utils.data as data
 11 | from meter import AverageMeter
 12 | from logger import Logger
 13 | # from video_transforms import *
 14 | from transforms import *
 15 | from Dataset import MyDataset
 16 | from models.p3d_model import P3D199, get_optim_policies
 17 | from models.C3D import C3D
 18 | from models.i3dpt import I3D
 19 | from utils import check_gpu, transfer_model, accuracy, get_learning_rate
 20 | from visualize import Visualizer
 21 | from lr_scheduler import CyclicLR
 22 | 
 23 | class Training(object):
 24 |     def __init__(self, name_list, num_classes=400, modality='RGB', **kwargs):
 25 |         self.__dict__.update(kwargs)
 26 |         self.num_classes = num_classes
 27 |         self.modality = modality
 28 |         self.name_list = name_list
 29 |         # set accuracy avg = 0
 30 |         self.count_early_stop = 0
 31 |         # Set best precision = 0
 32 |         self.best_prec1 = 0
 33 |         # init start epoch = 0
 34 |         self.start_epoch = 0
 35 | 
 36 |         if self.log_visualize != '':
 37 |             self.visualizer = Visualizer(logdir=self.log_visualize)
 38 | 
 39 |         self.checkDataFolder()
 40 | 
 41 |         self.loading_model()
 42 | 
 43 |         self.train_loader, self.val_loader = self.loading_data()
 44 | 
 45 |         # run
 46 |         self.processing()
 47 |         if self.random:
 48 |             print('random pick images')
 49 | 
 50 |     def check_early_stop(self, accuracy, logger, start_time):
 51 |         if self.best_prec1 <= accuracy:
 52 |             self.count_early_stop = 0
 53 |         else:
 54 |             self.count_early_stop += 1
 55 | 
 56 |         if self.count_early_stop > self.early_stop:
 57 |             print('Early stop')
 58 |             end_time = time.time()
 59 |             print("--- Total training time %s seconds ---" %
 60 |                   (end_time - start_time))
 61 |             logger.info("--- Total training time %s seconds ---" %
 62 |                         (end_time - start_time))
 63 |             exit()
 64 | 
 65 |     def checkDataFolder(self):
 66 |         try:
 67 |             os.stat('./' + self.model_type + '_' + self.data_set)
 68 |         except:
 69 |             os.mkdir('./' + self.model_type + '_' + self.data_set)
 70 |         self.data_folder = './' + self.model_type + '_' + self.data_set
 71 | 
 72 |     # Loading P3D model
 73 |     def loading_model(self):
 74 | 
 75 |         print('Loading %s model' % (self.model_type))
 76 | 
 77 |         if self.model_type == 'C3D':
 78 |             self.model = C3D()
 79 |             if self.pretrained:
 80 |                 self.model.load_state_dict(torch.load('c3d.pickle'))
 81 |         elif self.model_type == 'I3D':
 82 |             if self.pretrained:
 83 |                 self.model = I3D(num_classes=400, modality='rgb')
 84 |                 self.model.load_state_dict(
 85 |                     torch.load('kinetics_i3d_model_rgb.pth'))
 86 |             else:
 87 |                 self.model = I3D(num_classes=self.num_classes, modality='rgb')
 88 |         else:
 89 |             if self.pretrained:
 90 |                 print("=> using pre-trained model")
 91 |                 self.model = P3D199(
 92 |                     pretrained=True, num_classes=400, dropout=self.dropout)
 93 | 
 94 |             else:
 95 |                 print("=> creating model P3D")
 96 |                 self.model = P3D199(
 97 |                     pretrained=False, num_classes=400, dropout=self.dropout)
 98 |         # Transfer classes
 99 |         self.model = transfer_model(model=self.model, model_type=self.model_type, num_classes=self.num_classes)
100 | 
101 |         # Check gpu and run parallel
102 |         if check_gpu() > 0:
103 |             self.model = torch.nn.DataParallel(self.model).cuda()
104 | 
105 |         # define loss function (criterion) and optimizer
106 |         self.criterion = nn.CrossEntropyLoss()
107 | 
108 |         if check_gpu() > 0:
109 |             self.criterion = nn.CrossEntropyLoss().cuda()
110 | 
111 |         params = self.model.parameters()
112 |         if self.model_type == 'P3D':
113 |             params = get_optim_policies( model=self.model, modality=self.modality, enable_pbn=True)
114 | 
115 |         self.optimizer = optim.SGD(params=params, lr=self.lr, momentum=self.momentum, weight_decay=self.weight_decay)
116 | 
117 |         # self.scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer=self.optimizer, mode='min', patience=10, verbose=True)
118 | 
119 |         # optionally resume from a checkpoint
120 |         if self.resume:
121 |             if os.path.isfile(self.resume):
122 |                 print("=> loading checkpoint '{}'".format(self.resume))
123 |                 checkpoint = torch.load(self.resume)
124 |                 self.start_epoch = checkpoint['epoch']
125 |                 self.best_prec1 = checkpoint['best_prec1']
126 |                 self.model.load_state_dict(checkpoint['state_dict'])
127 |                 self.optimizer.load_state_dict(checkpoint['optimizer'])
128 |                 print("=> loaded checkpoint '{}' (epoch {})".format(
129 |                     self.evaluate, checkpoint['epoch']))
130 |             else:
131 |                 print("=> no checkpoint found at '{}'".format(self.resume))
132 | 
133 |         if self.evaluate:
134 |             file_model_best = os.path.join(
135 |                 self.data_folder, 'model_best.pth.tar')
136 |             if os.path.isfile(file_model_best):
137 |                 print("=> loading checkpoint '{}'".format('model_best.pth.tar'))
138 |                 checkpoint = torch.load(file_model_best)
139 |                 self.start_epoch = checkpoint['epoch']
140 |                 self.best_prec1 = checkpoint['best_prec1']
141 |                 self.model.load_state_dict(checkpoint['state_dict'])
142 |                 self.optimizer.load_state_dict(checkpoint['optimizer'])
143 |                 print("=> loaded checkpoint '{}' (epoch {})".format(
144 |                     self.evaluate, checkpoint['epoch']))
145 |             else:
146 |                 print("=> no checkpoint found at '{}'".format(self.resume))
147 | 
148 |         cudnn.benchmark = True
149 | 
150 |     # Loading data
151 |     def loading_data(self):
152 |         random = True if self.random else False
153 |         size = 160
154 |         if self.model_type == 'C3D':
155 |             size = 112
156 |         if self.model_type == 'I3D':
157 |             size = 224
158 | 
159 |         normalize = Normalize(mean=[0.485, 0.456, 0.406], std=[
160 |             0.229, 0.224, 0.225])
161 |         train_transformations = Compose([
162 |             RandomSizedCrop(size),
163 |             RandomHorizontalFlip(),
164 |             # Resize((size, size)),
165 |             # ColorJitter(
166 |             #     brightness=0.4,
167 |             #     contrast=0.4,
168 |             #     saturation=0.4,
169 |             # ),
170 |             ToTensor(),
171 |             normalize])
172 | 
173 |         val_transformations = Compose([
174 |             # Resize((182, 242)),
175 |             Resize(256),
176 |             CenterCrop(size),
177 |             ToTensor(),
178 |             normalize
179 |         ])
180 | 
181 |         train_dataset = MyDataset(
182 |             self.data,
183 |             data_folder="train",
184 |             name_list=self.name_list,
185 |             version="1",
186 |             transform=train_transformations,
187 |             num_frames=self.num_frames,
188 |             random=random
189 |         )
190 | 
191 |         val_dataset = MyDataset(
192 |             self.data,
193 |             data_folder="validation",
194 |             name_list=self.name_list,
195 |             version="1",
196 |             transform=val_transformations,
197 |             num_frames=self.num_frames,
198 |             random=random
199 |         )
200 | 
201 |         train_loader = data.DataLoader(
202 |             train_dataset,
203 |             batch_size=self.batch_size,
204 |             shuffle=True,
205 |             num_workers=self.workers,
206 |             pin_memory=True)
207 | 
208 |         val_loader = data.DataLoader(
209 |             val_dataset,
210 |             batch_size=self.batch_size,
211 |             shuffle=False,
212 |             num_workers=self.workers,
213 |             pin_memory=False)
214 | 
215 |         return (train_loader, val_loader)
216 | 
217 |     def processing(self):
218 |         log_file = os.path.join(self.data_folder, 'train.log')
219 | 
220 |         logger = Logger('train', log_file)
221 | 
222 |         iters = len(self.train_loader)
223 |         step_size = iters * 2
224 |         self.scheduler = CyclicLR(optimizer=self.optimizer, step_size=step_size, base_lr=self.lr)
225 | 
226 |         if self.evaluate:
227 |             self.validate(logger)
228 |             return
229 | 
230 |         iter_per_epoch = len(self.train_loader)
231 |         logger.info('Iterations per epoch: {0}'.format(iter_per_epoch))
232 |         print('Iterations per epoch: {0}'.format(iter_per_epoch))
233 | 
234 |         start_time = time.time()
235 | 
236 |         for epoch in range(self.start_epoch, self.epochs):
237 |             # self.adjust_learning_rate(epoch)
238 | 
239 |             # train for one epoch
240 |             train_losses, train_acc = self.train(logger, epoch)
241 | 
242 |             # evaluate on validation set
243 |             with torch.no_grad():
244 |                 val_losses, val_acc = self.validate(logger)
245 | 
246 |             # self.scheduler.step(val_losses.avg)
247 |             # log visualize
248 |             info_acc = {'train_acc': train_acc.avg, 'val_acc': val_acc.avg}
249 |             info_loss = {'train_loss': train_losses.avg, 'val_loss': val_losses.avg}
250 |             self.visualizer.write_summary(info_acc, info_loss, epoch + 1)
251 | 
252 |             self.visualizer.write_histogram(model=self.model, step=epoch + 1)
253 | 
254 |             # remember best Accuracy and save checkpoint
255 |             is_best = val_acc.avg > self.best_prec1
256 |             self.best_prec1 = max(val_acc.avg, self.best_prec1)
257 |             self.save_checkpoint({
258 |                 'epoch': epoch + 1,
259 |                 'state_dict': self.model.state_dict(),
260 |                 'best_prec1': self.best_prec1,
261 |                 'optimizer': self.optimizer.state_dict(),
262 |             }, is_best)
263 | 
264 |             self.check_early_stop(val_acc.avg, logger, start_time)
265 | 
266 |         end_time = time.time()
267 |         print("--- Total training time %s seconds ---" %
268 |               (end_time - start_time))
269 |         logger.info("--- Total training time %s seconds ---" %
270 |                     (end_time - start_time))
271 |         self.visualizer.writer_close()
272 | 
273 |     # Training
274 |     def train(self, logger, epoch):
275 |         batch_time = AverageMeter()
276 |         data_time = AverageMeter()
277 |         losses = AverageMeter()
278 |         acc = AverageMeter()
279 |         top1 = AverageMeter()
280 |         top5 = AverageMeter()
281 | 
282 |         rate = get_learning_rate(self.optimizer)[0]
283 |         # switch to train mode
284 |         self.model.train()
285 | 
286 |         end = time.time()
287 |         for i, (images, target) in enumerate(self.train_loader):
288 |             # adjust learning rate scheduler step
289 |             self.scheduler.batch_step()
290 | 
291 |             # measure data loading time
292 |             data_time.update(time.time() - end)
293 |             if check_gpu() > 0:
294 |                 images = images.cuda(async=True)
295 |                 target = target.cuda(async=True)
296 |             image_var = torch.autograd.Variable(images)
297 |             label_var = torch.autograd.Variable(target)
298 | 
299 |             self.optimizer.zero_grad()
300 | 
301 |             # compute y_pred
302 |             y_pred = self.model(image_var)
303 |             if self.model_type == 'I3D':
304 |                 y_pred = y_pred[0]
305 | 
306 |             loss = self.criterion(y_pred, label_var)
307 |             # measure accuracy and record loss
308 |             prec1, prec5 = accuracy(y_pred.data, target, topk=(1, 5))
309 |             losses.update(loss.item(), images.size(0))
310 |             acc.update(prec1.item(), images.size(0))
311 |             top1.update(prec1.item(), images.size(0))
312 |             top5.update(prec5.item(), images.size(0))
313 |             # compute gradient and do SGD step
314 | 
315 |             loss.backward()
316 |             self.optimizer.step()
317 | 
318 |             # measure elapsed time
319 |             batch_time.update(time.time() - end)
320 |             end = time.time()
321 | 
322 |             if i % self.print_freq == 0:
323 |                 print('Epoch: [{0}/{1}][{2}/{3}]\t'
324 |                       'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
325 |                       'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
326 |                       'Lr {rate:.5f}\t'
327 |                       'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
328 |                       'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
329 |                       'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(epoch, self.epochs, i, len(self.train_loader),
330 |                                                                       batch_time=batch_time, data_time=data_time,
331 |                                                                       rate=rate,
332 |                                                                       loss=losses, top1=top1, top5=top5))
333 | 
334 |         logger.info('Epoch: [{0}/{1}]\t'
335 |                     'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
336 |                     'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
337 |                     'Lr {rate:.5f}\t'
338 |                     'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
339 |                     'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
340 |                     'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(epoch, self.epochs, batch_time=batch_time,
341 |                                                                     data_time=data_time, rate=rate, loss=losses,
342 |                                                                     top1=top1,
343 |                                                                     top5=top5))
344 |         return losses, acc
345 | 
346 |     # Validation
347 |     def validate(self, logger):
348 |         batch_time = AverageMeter()
349 |         losses = AverageMeter()
350 |         acc = AverageMeter()
351 |         top1 = AverageMeter()
352 |         top5 = AverageMeter()
353 |         # switch to evaluate mode
354 |         self.model.eval()
355 | 
356 |         end = time.time()
357 |         for i, (images, labels) in enumerate(self.val_loader):
358 |             if check_gpu() > 0:
359 |                 images = images.cuda(async=True)
360 |                 labels = labels.cuda(async=True)
361 | 
362 |             image_var = torch.autograd.Variable(images)
363 |             label_var = torch.autograd.Variable(labels)
364 | 
365 |             # compute y_pred
366 |             y_pred = self.model(image_var)
367 |             if self.model_type == 'I3D':
368 |                 y_pred = y_pred[0]
369 | 
370 |             loss = self.criterion(y_pred, label_var)
371 | 
372 |             # measure accuracy and record loss
373 |             prec1, prec5 = accuracy(y_pred.data, labels, topk=(1, 5))
374 |             losses.update(loss.item(), images.size(0))
375 |             acc.update(prec1.item(), images.size(0))
376 |             top1.update(prec1.item(), images.size(0))
377 |             top5.update(prec5.item(), images.size(0))
378 |             # measure elapsed time
379 |             batch_time.update(time.time() - end)
380 |             end = time.time()
381 | 
382 |             if i % self.print_freq == 0:
383 |                 print('TrainVal: [{0}/{1}]\t'
384 |                       'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
385 |                       'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
386 |                       'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
387 |                       'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
388 |                     i, len(self.val_loader), batch_time=batch_time, loss=losses, top1=top1, top5=top5))
389 | 
390 |         print(
391 |             ' * Accuracy {acc.avg:.3f}  Loss {loss.avg:.3f}'.format(acc=acc, loss=losses))
392 |         logger.info(
393 |             ' * Accuracy {acc.avg:.3f}  Loss {loss.avg:.3f}'.format(acc=acc, loss=losses))
394 | 
395 |         return losses, acc
396 | 
397 |     # save checkpoint to file
398 |     def save_checkpoint(self, state, is_best):
399 |         checkpoint = os.path.join(self.data_folder, 'checkpoint.pth.tar')
400 |         torch.save(state, checkpoint)
401 |         model_best = os.path.join(self.data_folder, 'model_best.pth.tar')
402 |         if is_best:
403 |             shutil.copyfile(checkpoint, model_best)
404 | 
405 |     # adjust learning rate for each epoch
406 |     def adjust_learning_rate(self, epoch):
407 |         """Sets the learning rate to the initial LR decayed by 10 every 3K iterations"""
408 |         iters = len(self.train_loader)
409 |         num_epochs = 3000 // iters
410 |         decay = 0.1 ** (epoch // num_epochs)
411 |         lr = self.lr * decay
412 |         for param_group in self.optimizer.param_groups:
413 |             param_group['lr'] = lr * param_group['lr_mult']
414 |             param_group['weight_decay'] = decay * param_group['decay_mult']
415 | 


--------------------------------------------------------------------------------
/transforms.py:
--------------------------------------------------------------------------------
  1 | """
  2 | https://github.com/yjxiong/tsn-pytorch/blob/master/transforms.py
  3 | """
  4 | 
  5 | import torchvision
  6 | import random
  7 | from PIL import Image, ImageOps
  8 | import numpy as np
  9 | import numbers
 10 | import math
 11 | import torch
 12 | 
 13 | 
 14 | class Compose(object):
 15 |     def __init__(self, transforms):
 16 |         self.transforms = transforms
 17 | 
 18 |     def __call__(self, clip):
 19 |         for t in self.transforms:
 20 |             clip = t(clip)
 21 |         return clip
 22 | 
 23 | class ColorJitter(object):
 24 |     def __init__(self, brightness=0, contrast=0, saturation=0, hue=0):
 25 |         self.worker = torchvision.transforms.ColorJitter(brightness, contrast, saturation, hue)
 26 | 
 27 |     def __call__(self, img_group):
 28 |         return [self.worker(img) for img in img_group]
 29 | 
 30 | 
 31 | class RandomCrop(object):
 32 |     def __init__(self, size):
 33 |         if isinstance(size, numbers.Number):
 34 |             self.size = (int(size), int(size))
 35 |         else:
 36 |             self.size = size
 37 | 
 38 |     def __call__(self, img_group):
 39 | 
 40 |         w, h = img_group[0].size
 41 |         th, tw = self.size
 42 | 
 43 |         out_images = list()
 44 | 
 45 |         x1 = random.randint(0, w - tw)
 46 |         y1 = random.randint(0, h - th)
 47 | 
 48 |         for img in img_group:
 49 |             assert (img.size[0] == w and img.size[1] == h)
 50 |             if w == tw and h == th:
 51 |                 out_images.append(img)
 52 |             else:
 53 |                 out_images.append(img.crop((x1, y1, x1 + tw, y1 + th)))
 54 | 
 55 |         return out_images
 56 | 
 57 | 
 58 | class CenterCrop(object):
 59 |     def __init__(self, size):
 60 |         self.worker = torchvision.transforms.CenterCrop(size)
 61 | 
 62 |     def __call__(self, img_group):
 63 |         return [self.worker(img) for img in img_group]
 64 | 
 65 | 
 66 | class RandomHorizontalFlip(object):
 67 |     """Randomly horizontally flips the given PIL.Image with a probability of 0.5
 68 |     """
 69 | 
 70 |     def __init__(self, is_flow=False):
 71 |         self.is_flow = is_flow
 72 | 
 73 |     def __call__(self, img_group, is_flow=False):
 74 |         v = random.random()
 75 |         if v < 0.5:
 76 |             ret = [img.transpose(Image.FLIP_LEFT_RIGHT) for img in img_group]
 77 |             if self.is_flow:
 78 |                 for i in range(0, len(ret), 2):
 79 |                     ret[i] = ImageOps.invert(ret[i])  # invert flow pixel values when flipping
 80 |             return ret
 81 |         else:
 82 |             return img_group
 83 | 
 84 | 
 85 | class Normalize(object):
 86 |     def __init__(self, mean, std):
 87 |         self.mean = mean
 88 |         self.std = std
 89 | 
 90 |     def __call__(self, tensor):
 91 |         rep_mean = self.mean * (tensor.size()[0] // len(self.mean))
 92 |         rep_std = self.std * (tensor.size()[0] // len(self.std))
 93 |         # TODO: make efficient
 94 |         for t, m, s in zip(tensor, rep_mean, rep_std):
 95 |             t.sub_(m).div_(s)
 96 |         return tensor
 97 | 
 98 | 
 99 | class Scale(object):
100 |     """ Rescales the input PIL.Image to the given 'size'.
101 |     'size' will be the size of the smaller edge.
102 |     For example, if height > width, then image will be
103 |     rescaled to (size * height / width, size)
104 |     size: size of the smaller edge
105 |     interpolation: Default: PIL.Image.BILINEAR
106 |     """
107 | 
108 |     def __init__(self, size, interpolation=Image.BILINEAR):
109 |         self.worker = torchvision.transforms.Scale(size, interpolation)
110 | 
111 |     def __call__(self, img_group):
112 |         return [self.worker(img) for img in img_group]
113 | 
114 | 
115 | class Resize(object):
116 |     def __init__(self, size, interpolation=Image.BILINEAR):
117 |         self.worker = torchvision.transforms.Resize(size, interpolation)
118 | 
119 |     def __call__(self, img_group):
120 |         return [self.worker(img) for img in img_group]
121 | 
122 | 
123 | class OverSample(object):
124 |     def __init__(self, crop_size, scale_size=None):
125 |         self.crop_size = crop_size if not isinstance(crop_size, int) else (crop_size, crop_size)
126 | 
127 |         if scale_size is not None:
128 |             self.scale_worker = Scale(scale_size)
129 |         else:
130 |             self.scale_worker = None
131 | 
132 |     def __call__(self, img_group):
133 | 
134 |         if self.scale_worker is not None:
135 |             img_group = self.scale_worker(img_group)
136 | 
137 |         image_w, image_h = img_group[0].size
138 |         crop_w, crop_h = self.crop_size
139 | 
140 |         offsets = MultiScaleCrop.fill_fix_offset(False, image_w, image_h, crop_w, crop_h)
141 |         oversample_group = list()
142 |         for o_w, o_h in offsets:
143 |             normal_group = list()
144 |             flip_group = list()
145 |             for i, img in enumerate(img_group):
146 |                 crop = img.crop((o_w, o_h, o_w + crop_w, o_h + crop_h))
147 |                 normal_group.append(crop)
148 |                 flip_crop = crop.copy().transpose(Image.FLIP_LEFT_RIGHT)
149 | 
150 |                 if img.mode == 'L' and i % 2 == 0:
151 |                     flip_group.append(ImageOps.invert(flip_crop))
152 |                 else:
153 |                     flip_group.append(flip_crop)
154 | 
155 |             oversample_group.extend(normal_group)
156 |             oversample_group.extend(flip_group)
157 |         return oversample_group
158 | 
159 | 
160 | class MultiScaleCrop(object):
161 | 
162 |     def __init__(self, input_size, scales=None, max_distort=1, fix_crop=True, more_fix_crop=True):
163 |         self.scales = scales if scales is not None else [1, .875, .75, .66]
164 |         self.max_distort = max_distort
165 |         self.fix_crop = fix_crop
166 |         self.more_fix_crop = more_fix_crop
167 |         self.input_size = input_size if not isinstance(input_size, int) else [input_size, input_size]
168 |         self.interpolation = Image.BILINEAR
169 | 
170 |     def __call__(self, img_group):
171 | 
172 |         im_size = img_group[0].size
173 | 
174 |         crop_w, crop_h, offset_w, offset_h = self._sample_crop_size(im_size)
175 |         crop_img_group = [img.crop((offset_w, offset_h, offset_w + crop_w, offset_h + crop_h)) for img in img_group]
176 |         ret_img_group = [img.resize((self.input_size[0], self.input_size[1]), self.interpolation)
177 |                          for img in crop_img_group]
178 |         return ret_img_group
179 | 
180 |     def _sample_crop_size(self, im_size):
181 |         image_w, image_h = im_size[0], im_size[1]
182 | 
183 |         # find a crop size
184 |         base_size = min(image_w, image_h)
185 |         crop_sizes = [int(base_size * x) for x in self.scales]
186 |         crop_h = [self.input_size[1] if abs(x - self.input_size[1]) < 3 else x for x in crop_sizes]
187 |         crop_w = [self.input_size[0] if abs(x - self.input_size[0]) < 3 else x for x in crop_sizes]
188 | 
189 |         pairs = []
190 |         for i, h in enumerate(crop_h):
191 |             for j, w in enumerate(crop_w):
192 |                 if abs(i - j) <= self.max_distort:
193 |                     pairs.append((w, h))
194 | 
195 |         crop_pair = random.choice(pairs)
196 |         if not self.fix_crop:
197 |             w_offset = random.randint(0, image_w - crop_pair[0])
198 |             h_offset = random.randint(0, image_h - crop_pair[1])
199 |         else:
200 |             w_offset, h_offset = self._sample_fix_offset(image_w, image_h, crop_pair[0], crop_pair[1])
201 | 
202 |         return crop_pair[0], crop_pair[1], w_offset, h_offset
203 | 
204 |     def _sample_fix_offset(self, image_w, image_h, crop_w, crop_h):
205 |         offsets = self.fill_fix_offset(self.more_fix_crop, image_w, image_h, crop_w, crop_h)
206 |         return random.choice(offsets)
207 | 
208 |     @staticmethod
209 |     def fill_fix_offset(more_fix_crop, image_w, image_h, crop_w, crop_h):
210 |         w_step = (image_w - crop_w) // 4
211 |         h_step = (image_h - crop_h) // 4
212 | 
213 |         ret = list()
214 |         ret.append((0, 0))  # upper left
215 |         ret.append((4 * w_step, 0))  # upper right
216 |         ret.append((0, 4 * h_step))  # lower left
217 |         ret.append((4 * w_step, 4 * h_step))  # lower right
218 |         ret.append((2 * w_step, 2 * h_step))  # center
219 | 
220 |         if more_fix_crop:
221 |             ret.append((0, 2 * h_step))  # center left
222 |             ret.append((4 * w_step, 2 * h_step))  # center right
223 |             ret.append((2 * w_step, 4 * h_step))  # lower center
224 |             ret.append((2 * w_step, 0 * h_step))  # upper center
225 | 
226 |             ret.append((1 * w_step, 1 * h_step))  # upper left quarter
227 |             ret.append((3 * w_step, 1 * h_step))  # upper right quarter
228 |             ret.append((1 * w_step, 3 * h_step))  # lower left quarter
229 |             ret.append((3 * w_step, 3 * h_step))  # lower righ quarter
230 | 
231 |         return ret
232 | 
233 | 
234 | class RandomSizedCrop(object):
235 |     """Random crop the given PIL.Image to a random size of (0.08 to 1.0) of the original size
236 |     and and a random aspect ratio of 3/4 to 4/3 of the original aspect ratio
237 |     This is popularly used to train the Inception networks
238 |     size: size of the smaller edge
239 |     interpolation: Default: PIL.Image.BILINEAR
240 |     """
241 | 
242 |     def __init__(self, size, interpolation=Image.BILINEAR):
243 |         self.size = size
244 |         self.interpolation = interpolation
245 | 
246 |     def __call__(self, img_group):
247 |         for attempt in range(10):
248 |             area = img_group[0].size[0] * img_group[0].size[1]
249 |             target_area = random.uniform(0.08, 1.0) * area
250 |             aspect_ratio = random.uniform(3. / 4, 4. / 3)
251 | 
252 |             w = int(round(math.sqrt(target_area * aspect_ratio)))
253 |             h = int(round(math.sqrt(target_area / aspect_ratio)))
254 | 
255 |             if random.random() < 0.5:
256 |                 w, h = h, w
257 | 
258 |             if w <= img_group[0].size[0] and h <= img_group[0].size[1]:
259 |                 x1 = random.randint(0, img_group[0].size[0] - w)
260 |                 y1 = random.randint(0, img_group[0].size[1] - h)
261 |                 found = True
262 |                 break
263 |         else:
264 |             found = False
265 |             x1 = 0
266 |             y1 = 0
267 | 
268 |         if found:
269 |             out_group = list()
270 |             for img in img_group:
271 |                 img = img.crop((x1, y1, x1 + w, y1 + h))
272 |                 assert (img.size == (w, h))
273 |                 out_group.append(img.resize((self.size, self.size), self.interpolation))
274 |             return out_group
275 |         else:
276 |             # Fallback
277 |             scale = Scale(self.size, interpolation=self.interpolation)
278 |             crop = RandomCrop(self.size)
279 |             return crop(scale(img_group))
280 | 
281 | 
282 | class Stack(object):
283 | 
284 |     def __init__(self, roll=False):
285 |         self.roll = roll
286 | 
287 |     def __call__(self, img_group):
288 |         if img_group[0].mode == 'L':
289 |             return np.concatenate([np.expand_dims(x, 2) for i, x in enumerate(img_group)], axis=2)
290 |         elif img_group[0].mode == 'RGB':
291 |             if self.roll:
292 |                 return np.concatenate([np.array(x)[:, :, ::-1] for i, x in enumerate(img_group)], axis=2)
293 |             else:
294 |                 return np.concatenate(img_group, axis=2)
295 | 
296 | 
297 | # class ToTensor2(object):
298 | #     """ Converts a PIL.Image (RGB) or numpy.ndarray (H x W x C) in the range [0, 255]
299 | #     to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 1.0] """
300 | #
301 | #     def __init__(self, channel_nb=3, div=True, numpy=False):
302 | #         self.div = div
303 | #         self.channel_nb=channel_nb
304 | #         self.numpy = numpy
305 | #
306 | #     def __call__(self, pic):
307 | #         if isinstance(pic, np.ndarray):
308 | #             # handle numpy array
309 | #             img = torch.from_numpy(pic).permute(2, 0, 1).contiguous()
310 | #             # print(img.size())
311 | #         else:
312 | #             # handle PIL Image
313 | #             img = torch.ByteTensor(torch.ByteStorage.from_buffer(pic.tobytes()))
314 | #             img = img.view(pic.size[1], pic.size[0], len(pic.mode))
315 | #             # put it from HWC to CHW format
316 | #             # yikes, this transpose takes 80% of the loading time/CPU
317 | #             img = img.transpose(0, 1).transpose(0, 2).contiguous()
318 | #
319 | #         return img.float().div(255) if self.div else img.float()
320 | 
321 | class ToTensor(object):
322 |     """Convert a list of m (H x W x C) numpy.ndarrays in the range [0, 255]
323 |     to a torch.FloatTensor of shape (C x m x H x W) in the range [0, 1.0]
324 |     """
325 | 
326 |     def __init__(self, channel_nb=3, div_255=True, numpy=False):
327 |         self.channel_nb = channel_nb
328 |         self.div_255 = div_255
329 |         self.numpy = numpy
330 | 
331 |     def convert_img(self, img):
332 |         """Converts (H, W, C) numpy.ndarray to (C, W, H) format
333 |         """
334 |         if len(img.shape) == 3:
335 |             img = img.transpose(2, 0, 1)
336 |         if len(img.shape) == 2:
337 |             img = np.expand_dims(img, 0)
338 |         return img
339 | 
340 |     def __call__(self, clip):
341 |         """
342 |         Args: clip (list of numpy.ndarray): clip (list of images)
343 |         to be converted to tensor.
344 |         """
345 | 
346 |         h, w = clip[0].size
347 |         np_clip = np.zeros([self.channel_nb, len(clip), int(h), int(w)])
348 | 
349 |         # Convert
350 |         for img_idx, img in enumerate(clip):
351 |             if isinstance(img, np.ndarray):
352 |                 pass
353 |             elif isinstance(img, Image.Image):
354 |                 img = np.array(img, copy=False)
355 |             else:
356 |                 raise TypeError('Expected numpy.ndarray or PIL.Image\
357 |                         but got list of {0}'.format(type(clip[0])))
358 |             img = self.convert_img(img)
359 |             np_clip[:, img_idx, :, :] = img
360 |         if self.numpy:
361 |             if self.div_255:
362 |                 np_clip = np_clip / 255
363 |             return np_clip
364 | 
365 |         else:
366 |             tensor_clip = torch.from_numpy(np_clip)
367 | 
368 |             if not isinstance(tensor_clip, torch.FloatTensor):
369 |                 tensor_clip = tensor_clip.float()
370 |             if self.div_255:
371 |                 tensor_clip = tensor_clip.div(255)
372 |             return tensor_clip
373 | 
374 | class IdentityTransform(object):
375 | 
376 |     def __call__(self, data):
377 |         return data
378 | 
379 | 
380 | if __name__ == "__main__":
381 |     trans = Compose([
382 |         Scale(256),
383 |         RandomCrop(224),
384 |         Stack(),
385 |         ToTensor(),
386 |         Normalize(
387 |             mean=[.485, .456, .406],
388 |             std=[.229, .224, .225]
389 |         )]
390 |     )
391 | 
392 |     im = Image.open('dog.0.jpg')
393 | 
394 |     color_group = [im] * 3
395 |     rst = trans(color_group)
396 | 
397 |     gray_group = [im.convert('L')] * 9
398 |     gray_rst = trans(gray_group)
399 | 
400 |     trans2 = torchvision.transforms.Compose([
401 |         RandomSizedCrop(256),
402 |         Stack(),
403 |         ToTensor(),
404 |         Normalize(
405 |             mean=[.485, .456, .406],
406 |             std=[.229, .224, .225])
407 |     ])
408 |     print(trans2(color_group))
409 | 


--------------------------------------------------------------------------------
/utils.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import torch.nn as nn
 3 | from models.i3dpt import Unit3Dpy, I3D
 4 | 
 5 | 
 6 | # get accuracy from y pred
 7 | def accuracy(y_pred, y_actual, topk=(1,)):
 8 |     """Computes the precision@k for the specified values of k"""
 9 |     maxk = max(topk)
10 |     batch_size = y_actual.size(0)
11 | 
12 |     _, pred = y_pred.topk(maxk, 1, True, True)
13 |     pred = pred.t()
14 |     correct = pred.eq(y_actual.view(1, -1).expand_as(pred))
15 | 
16 |     res = []
17 |     for k in topk:
18 |         correct_k = correct[:k].view(-1).float().sum(0)
19 |         res.append(correct_k.mul_(100.0 / batch_size))
20 | 
21 |     return res
22 | 
23 | 
24 | def check_gpu():
25 |     num_gpus = 0
26 |     if torch.cuda.is_available():
27 |         num_gpus = torch.cuda.device_count()
28 |     return num_gpus
29 | 
30 | 
31 | def transfer_model(model, model_type='P3D', num_classes=101):
32 |     if model_type == 'P3D':
33 |         num_ftrs = model.fc.in_features
34 |         model.fc = nn.Linear(num_ftrs, num_classes)
35 |     elif model_type == 'C3D':
36 |         model.fc8 = nn.Linear(4096, num_classes)
37 |     elif model_type == "I3D":
38 |         conv3d_0c_1x1 = Unit3Dpy(
39 |             in_channels=1024,
40 |             out_channels=num_classes,
41 |             kernel_size=(1, 1, 1),
42 |             activation=None,
43 |             use_bias=True,
44 |             use_bn=False)
45 |         model.conv3d_0c_1x1 = conv3d_0c_1x1
46 |     return model
47 | 
48 | 
49 | def get_learning_rate(optimizer):
50 |     lr = []
51 |     for param_group in optimizer.param_groups:
52 |         lr += [param_group['lr']]
53 |     return lr


--------------------------------------------------------------------------------
/video_transforms.py:
--------------------------------------------------------------------------------
  1 | import torch
  2 | import torchvision
  3 | import math
  4 | import random
  5 | import numpy as np
  6 | from PIL import Image, ImageOps, ImageEnhance
  7 | 
  8 | 
  9 | class Compose(object):
 10 |     def __init__(self, transforms):
 11 |         self.transforms = transforms
 12 | 
 13 |     def __call__(self, clip):
 14 |         for t in self.transforms:
 15 |             clip = t(clip)
 16 |         return clip
 17 | 
 18 | 
 19 | class Resize(object):
 20 |     def __init__(self, size, interpolation=Image.BILINEAR):
 21 |         self.worker = torchvision.transforms.Resize(size, interpolation)
 22 | 
 23 |     def __call__(self, img_group):
 24 |         return [self.worker(img) for img in img_group]
 25 | 
 26 | 
 27 | class CenterCrop(object):
 28 |     def __init__(self, size):
 29 |         self.worker = torchvision.transforms.CenterCrop(size)
 30 | 
 31 |     def __call__(self, img_group):
 32 |         return [self.worker(img) for img in img_group]
 33 | 
 34 | 
 35 | class RandomHorizontalFlip(object):
 36 |     def __init__(self, is_flow=False):
 37 |         self.is_flow = is_flow
 38 | 
 39 |     def __call__(self, img_group, is_flow=False):
 40 |         v = random.random()
 41 |         if v < 0.5:
 42 |             ret = [img.transpose(Image.FLIP_LEFT_RIGHT) for img in img_group]
 43 |             if self.is_flow:
 44 |                 for i in range(0, len(ret), 2):
 45 |                     ret[i] = ImageOps.invert(ret[i])
 46 |             return ret
 47 |         else:
 48 |             return img_group
 49 | 
 50 | 
 51 | class RandomResizedCrop(object):
 52 |     def __init__(self, size, interpolation=Image.BILINEAR):
 53 |         self.size = size
 54 |         self.interpolation = interpolation
 55 | 
 56 |     def __call__(self, img_group):
 57 |         for attempt in range(10):
 58 |             area = img_group[0].size[0] * img_group[0].size[1]
 59 |             target_area = random.uniform(0.08, 1.0) * area
 60 |             aspect_ratio = random.uniform(3. / 4, 4. / 3)
 61 | 
 62 |             w = int(round(math.sqrt(target_area * aspect_ratio)))
 63 |             h = int(round(math.sqrt(target_area / aspect_ratio)))
 64 | 
 65 |             if random.random() < 0.5:
 66 |                 w, h = h, w
 67 | 
 68 |             if w <= img_group[0].size[0] and h <= img_group[0].size[1]:
 69 |                 x1 = random.randint(0, img_group[0].size[0] - w)
 70 |                 y1 = random.randint(0, img_group[0].size[1] - h)
 71 |                 found = True
 72 |                 break
 73 |         else:
 74 |             found = False
 75 |             x1 = 0
 76 |             y1 = 0
 77 | 
 78 |         if found:
 79 |             out_group = list()
 80 |             for img in img_group:
 81 |                 img = img.crop((x1, y1, x1 + w, y1 + h))
 82 |                 assert (img.size == (w, h))
 83 |                 out_group.append(img.resize((self.size, self.size), self.interpolation))
 84 |             return out_group
 85 |         else:
 86 |             scale = Resize(self.size, interpolation=self.interpolation)
 87 |             return scale(img_group)
 88 | 
 89 | 
 90 | class Normalize(object):
 91 |     def __init__(self, mean, std, channel=3, num_frames=16, size=160):
 92 |         self.mean = mean
 93 |         self.std = std
 94 |         self.channel = channel
 95 |         self.num_frames = num_frames
 96 |         self.size = size
 97 | 
 98 |     def __call__(self, tensor):
 99 |         rep_mean = self.mean * (tensor.size()[1])
100 |         rep_std = self.std * (tensor.size()[1])
101 |         tensor = tensor.contiguous().view(self.channel * self.num_frames, self.size, self.size)
102 |         for t, m, s in zip(tensor, rep_mean, rep_std):
103 |             t.sub_(m).div_(s)
104 |         tensor = tensor.view(self.channel, self.num_frames, self.size, self.size)
105 |         return tensor
106 | 
107 | 
108 | class ToTensor(object):
109 |     def __init__(self):
110 |         self.worker = torchvision.transforms.ToTensor()
111 | 
112 |     def __call__(self, img_group):
113 |         img_group = [self.worker(img) for img in img_group]
114 |         img_group = [img.numpy() for img in img_group]
115 |         img_group = np.array(img_group)
116 |         img_group = img_group.transpose(1, 0, 2, 3)
117 |         img_group = torch.from_numpy(img_group)
118 |         return img_group
119 | 


--------------------------------------------------------------------------------
/visualize.py:
--------------------------------------------------------------------------------
 1 | from tensorboardX import SummaryWriter
 2 | 
 3 | class Visualizer(object):
 4 |     def __init__(self, logdir='./runs'):
 5 |         self.writer = SummaryWriter(logdir)
 6 | 
 7 |     def write_graph(self, model, dummy_input):
 8 |         self.writer.add_graph(model, (dummy_input, ))
 9 | 
10 |     def write_summary(self, info_acc, info_loss, epoch):
11 |         self.writer.add_scalars('Accuracy', info_acc, epoch)
12 |         self.writer.add_scalars('Loss', info_loss, epoch)
13 | 
14 |     def write_histogram(self, model, step):
15 |         for name, param in model.named_parameters():
16 |             self.writer.add_histogram(name, param.clone().cpu().data.numpy(), step)
17 | 
18 |     def writer_close(self):
19 |         self.writer.export_scalars_to_json("./all_scalars.json")
20 |         self.writer.close()


--------------------------------------------------------------------------------