├── .DS_Store
├── .idea
├── .gitignore
├── .name
├── TorchProjects.iml
├── inspectionProfiles
│ ├── Project_Default.xml
│ └── profiles_settings.xml
├── jupyter-settings.xml
├── misc.xml
├── modules.xml
└── vcs.xml
├── LICENSE
├── Model
├── model01.pth
└── module.py
├── PaperReplicate
├── .DS_Store
├── Self_Attention_from_Scratch
│ ├── .DS_Store
│ ├── Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38.md
│ ├── Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38
│ │ ├── Untitled 1.png
│ │ ├── Untitled 2.png
│ │ ├── Untitled 3.png
│ │ ├── Untitled 4.png
│ │ ├── Untitled 5.png
│ │ ├── Untitled 6.png
│ │ └── Untitled.png
│ ├── self_attention.py
│ └── self_attention_mechanism.ipynb
├── VisionTransformer
│ └── replicate_vit.ipynb
├── data_setup.py
├── helper_functions.py
├── predictions.py
└── utils.py
├── PracticePytorch
├── .DS_Store
├── 02.Classification.ipynb
├── 03CV.ipynb
├── 04.CustomData.ipynb
├── 06.TransferLearning.ipynb
├── 07.ExperimentsTracking.ipynb
├── Custom_dataset_dataloader_transform.ipynb
├── backbone.py
├── c2.ipynb
├── faces
│ ├── 0805personali01.jpg
│ ├── 1084239450_e76e00b7e7.jpg
│ ├── 10comm-decarlo.jpg
│ ├── 110276240_bec305da91.jpg
│ ├── 1198_0_861.jpg
│ ├── 137341995_e7c48e9a75.jpg
│ ├── 1383023626_8a49e4879a.jpg
│ ├── 144044282_87cf3ff76e.jpg
│ ├── 152601997_ec6429a43c.jpg
│ ├── 1549040388_b99e9fa295.jpg
│ ├── 1878519279_f905d4f34e.jpg
│ ├── 2046713398_91aaa6fe1c.jpg
│ ├── 2173711035_dbd53b4f9f.jpg
│ ├── 2210514040_6b03ff2629.jpg
│ ├── 2322901504_08122b01ba.jpg
│ ├── 2327253037_66a61ea6fe.jpg
│ ├── 2328398005_d328a70b4c.jpg
│ ├── 2370961440_6bc8ce346c.jpg
│ ├── 2382SJ8.jpg
│ ├── 252418361_440b75751b.jpg
│ ├── 262007783_943bbcf613.jpg
│ ├── 2633371780_45b740b670.jpg
│ ├── 2647088981_60e9fe40cd.jpg
│ ├── 2711409561_a0786a3d3d.jpg
│ ├── 2722779845_7fcb64a096.jpg
│ ├── 2795838930_0cc5aa5f41.jpg
│ ├── 2902323565_100017b63c.jpg
│ ├── 2902760364_89c50bde40.jpg
│ ├── 2956581526_cd803f2daa.jpg
│ ├── 297448785_b2dda4b2c0.jpg
│ ├── 299733036_fff5ea6f8e.jpg
│ ├── 303808204_1f744bc407.jpg
│ ├── 3074791551_baee7fa0c1.jpg
│ ├── 3152653555_68322314f3.jpg
│ ├── 3264867945_fe18d442c1.jpg
│ ├── 3273658251_b95f65c244.jpg
│ ├── 3298715079_5af7c78fcb.jpg
│ ├── 3325611505_ddc7beffa1.jpg
│ ├── 3362762930_24f76cb89c.jpg
│ ├── 343583208_e986824d77.jpg
│ ├── 3461016494_56cce9c984.jpg
│ ├── 348272697_832ce65324.jpg
│ ├── 3534188114_2108895291.jpg
│ ├── 3534189272_8ef88ba368.jpg
│ ├── 3555944509_7b477069c6.jpg
│ ├── 3574737496_6ee8207045.jpg
│ ├── 362167809_d5a5dcbfdb.jpg
│ ├── 363149951_8be04dc6c0.jpg
│ ├── 3638950581_3387685d3a.jpg
│ ├── 3646828311_bfeb429ef7.jpg
│ ├── 3689162471_5f9ffb5aa0.jpg
│ ├── 3718903026_c1bf5dfcf8.jpg
│ ├── 3790616528_297c0ac935.jpg
│ ├── 3855944735_e252959937.jpg
│ ├── 3856149136_d4595ffdd4.jpg
│ ├── 3872768751_e60d7fdbd5.jpg
│ ├── 529447797_0f9d2fb756.jpg
│ ├── 57635685_d41c98f8ca.jpg
│ ├── 809285949_6889026b53.jpg
│ ├── 92053278_be61a225d2.jpg
│ ├── 96063776_bdb3617b64.jpg
│ ├── 97308305_4b737d0873.jpg
│ ├── britney-bald.jpg
│ ├── create_landmark_dataset.py
│ ├── deeny.peggy.jpg
│ ├── face_landmarks.csv
│ ├── matt-mathes.jpg
│ ├── person-7.jpg
│ ├── person.jpg
│ ├── person_TjahjonoDGondhowiardjo.jpg
│ └── personalpic.jpg
└── runs
│ ├── Jun18_14-46-04_Sherlocks-MacBook-Pro.local
│ └── events.out.tfevents.1687117564.Sherlocks-MacBook-Pro.local.51259.0
│ ├── Jun18_15-03-00_Sherlocks-MacBook-Pro.local
│ ├── events.out.tfevents.1687118580.Sherlocks-MacBook-Pro.local.51259.1
│ └── events.out.tfevents.1687119446.Sherlocks-MacBook-Pro.local.51259.2
│ └── efficientnet_b0_pizza_steak_sushi
│ ├── events.out.tfevents.1686965428.Sherlocks-MacBook-Pro.local.31548.0
│ ├── events.out.tfevents.1686965887.Sherlocks-MacBook-Pro.local.32194.0
│ └── events.out.tfevents.1686966251.Sherlocks-MacBook-Pro.local.32194.1
├── Python&Probability&Statistics
└── BasicPython
│ └── Numpy_Exercise.ipynb
├── README.md
├── Signal_Processing
├── .DS_Store
├── Intro
│ ├── .DS_Store
│ ├── glassDance.mat
│ ├── sigprocMXC_filterGlass.ipynb
│ └── sigprocMXC_filterGlass.m
├── SelectiveSearch
│ ├── Part 3 Object Detection with Pascal VOC2012 - Selective Search.ipynb
│ ├── __pycache__
│ │ └── slective_search.cpython-310.pyc
│ ├── image
│ │ ├── 2012_001297.jpg
│ │ ├── cats.png
│ │ ├── example_id1.JPG
│ │ ├── example_id2.JPG
│ │ ├── example_id3.JPG
│ │ ├── example_id4.JPG
│ │ └── example_image_easy.JPG
│ └── slective_search.py
└── TimeSeriesDenoising
│ ├── .DS_Store
│ ├── denoising_codeChallenge.mat
│ ├── emg4TKEO.mat
│ ├── eyedat.mat
│ ├── read_data.py
│ ├── sigprocMXC_timeSeriesDenoising.ipynb
│ └── templateProjection.mat
├── __pycache__
├── data_setup.cpython-310.pyc
├── data_setup.cpython-38.pyc
├── helper_functions.cpython-38.pyc
├── helper_functions.cpython-39.pyc
├── module.cpython-38.pyc
├── utils.cpython-310.pyc
└── utils.cpython-38.pyc
└── test.ipynb
/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/.DS_Store
--------------------------------------------------------------------------------
/.idea/.gitignore:
--------------------------------------------------------------------------------
1 | # Default ignored files
2 | /shelf/
3 | /workspace.xml
4 | # Editor-based HTTP Client requests
5 | /httpRequests/
6 | # Datasource local storage ignored files
7 | /dataSources/
8 | /dataSources.local.xml
9 |
--------------------------------------------------------------------------------
/.idea/.name:
--------------------------------------------------------------------------------
1 | sigprocMXC_filterGlass.ipynb
--------------------------------------------------------------------------------
/.idea/TorchProjects.iml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
--------------------------------------------------------------------------------
/.idea/inspectionProfiles/Project_Default.xml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
15 |
16 |
17 |
22 |
23 |
24 |
25 |
26 |
27 |
28 |
29 |
30 |
31 |
--------------------------------------------------------------------------------
/.idea/inspectionProfiles/profiles_settings.xml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
6 |
--------------------------------------------------------------------------------
/.idea/jupyter-settings.xml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
32 |
33 |
34 |
--------------------------------------------------------------------------------
/.idea/misc.xml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
--------------------------------------------------------------------------------
/.idea/modules.xml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
--------------------------------------------------------------------------------
/.idea/vcs.xml:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
6 |
--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
1 | GNU GENERAL PUBLIC LICENSE
2 | Version 3, 29 June 2007
3 |
4 | Copyright (C) 2007 Free Software Foundation, Inc.
5 | Everyone is permitted to copy and distribute verbatim copies
6 | of this license document, but changing it is not allowed.
7 |
8 | Preamble
9 |
10 | The GNU General Public License is a free, copyleft license for
11 | software and other kinds of works.
12 |
13 | The licenses for most software and other practical works are designed
14 | to take away your freedom to share and change the works. By contrast,
15 | the GNU General Public License is intended to guarantee your freedom to
16 | share and change all versions of a program--to make sure it remains free
17 | software for all its users. We, the Free Software Foundation, use the
18 | GNU General Public License for most of our software; it applies also to
19 | any other work released this way by its authors. You can apply it to
20 | your programs, too.
21 |
22 | When we speak of free software, we are referring to freedom, not
23 | price. Our General Public Licenses are designed to make sure that you
24 | have the freedom to distribute copies of free software (and charge for
25 | them if you wish), that you receive source code or can get it if you
26 | want it, that you can change the software or use pieces of it in new
27 | free programs, and that you know you can do these things.
28 |
29 | To protect your rights, we need to prevent others from denying you
30 | these rights or asking you to surrender the rights. Therefore, you have
31 | certain responsibilities if you distribute copies of the software, or if
32 | you modify it: responsibilities to respect the freedom of others.
33 |
34 | For example, if you distribute copies of such a program, whether
35 | gratis or for a fee, you must pass on to the recipients the same
36 | freedoms that you received. You must make sure that they, too, receive
37 | or can get the source code. And you must show them these terms so they
38 | know their rights.
39 |
40 | Developers that use the GNU GPL protect your rights with two steps:
41 | (1) assert copyright on the software, and (2) offer you this License
42 | giving you legal permission to copy, distribute and/or modify it.
43 |
44 | For the developers' and authors' protection, the GPL clearly explains
45 | that there is no warranty for this free software. For both users' and
46 | authors' sake, the GPL requires that modified versions be marked as
47 | changed, so that their problems will not be attributed erroneously to
48 | authors of previous versions.
49 |
50 | Some devices are designed to deny users access to install or run
51 | modified versions of the software inside them, although the manufacturer
52 | can do so. This is fundamentally incompatible with the aim of
53 | protecting users' freedom to change the software. The systematic
54 | pattern of such abuse occurs in the area of products for individuals to
55 | use, which is precisely where it is most unacceptable. Therefore, we
56 | have designed this version of the GPL to prohibit the practice for those
57 | products. If such problems arise substantially in other domains, we
58 | stand ready to extend this provision to those domains in future versions
59 | of the GPL, as needed to protect the freedom of users.
60 |
61 | Finally, every program is threatened constantly by software patents.
62 | States should not allow patents to restrict development and use of
63 | software on general-purpose computers, but in those that do, we wish to
64 | avoid the special danger that patents applied to a free program could
65 | make it effectively proprietary. To prevent this, the GPL assures that
66 | patents cannot be used to render the program non-free.
67 |
68 | The precise terms and conditions for copying, distribution and
69 | modification follow.
70 |
71 | TERMS AND CONDITIONS
72 |
73 | 0. Definitions.
74 |
75 | "This License" refers to version 3 of the GNU General Public License.
76 |
77 | "Copyright" also means copyright-like laws that apply to other kinds of
78 | works, such as semiconductor masks.
79 |
80 | "The Program" refers to any copyrightable work licensed under this
81 | License. Each licensee is addressed as "you". "Licensees" and
82 | "recipients" may be individuals or organizations.
83 |
84 | To "modify" a work means to copy from or adapt all or part of the work
85 | in a fashion requiring copyright permission, other than the making of an
86 | exact copy. The resulting work is called a "modified version" of the
87 | earlier work or a work "based on" the earlier work.
88 |
89 | A "covered work" means either the unmodified Program or a work based
90 | on the Program.
91 |
92 | To "propagate" a work means to do anything with it that, without
93 | permission, would make you directly or secondarily liable for
94 | infringement under applicable copyright law, except executing it on a
95 | computer or modifying a private copy. Propagation includes copying,
96 | distribution (with or without modification), making available to the
97 | public, and in some countries other activities as well.
98 |
99 | To "convey" a work means any kind of propagation that enables other
100 | parties to make or receive copies. Mere interaction with a user through
101 | a computer network, with no transfer of a copy, is not conveying.
102 |
103 | An interactive user interface displays "Appropriate Legal Notices"
104 | to the extent that it includes a convenient and prominently visible
105 | feature that (1) displays an appropriate copyright notice, and (2)
106 | tells the user that there is no warranty for the work (except to the
107 | extent that warranties are provided), that licensees may convey the
108 | work under this License, and how to view a copy of this License. If
109 | the interface presents a list of user commands or options, such as a
110 | menu, a prominent item in the list meets this criterion.
111 |
112 | 1. Source Code.
113 |
114 | The "source code" for a work means the preferred form of the work
115 | for making modifications to it. "Object code" means any non-source
116 | form of a work.
117 |
118 | A "Standard Interface" means an interface that either is an official
119 | standard defined by a recognized standards body, or, in the case of
120 | interfaces specified for a particular programming language, one that
121 | is widely used among developers working in that language.
122 |
123 | The "System Libraries" of an executable work include anything, other
124 | than the work as a whole, that (a) is included in the normal form of
125 | packaging a Major Component, but which is not part of that Major
126 | Component, and (b) serves only to enable use of the work with that
127 | Major Component, or to implement a Standard Interface for which an
128 | implementation is available to the public in source code form. A
129 | "Major Component", in this context, means a major essential component
130 | (kernel, window system, and so on) of the specific operating system
131 | (if any) on which the executable work runs, or a compiler used to
132 | produce the work, or an object code interpreter used to run it.
133 |
134 | The "Corresponding Source" for a work in object code form means all
135 | the source code needed to generate, install, and (for an executable
136 | work) run the object code and to modify the work, including scripts to
137 | control those activities. However, it does not include the work's
138 | System Libraries, or general-purpose tools or generally available free
139 | programs which are used unmodified in performing those activities but
140 | which are not part of the work. For example, Corresponding Source
141 | includes interface definition files associated with source files for
142 | the work, and the source code for shared libraries and dynamically
143 | linked subprograms that the work is specifically designed to require,
144 | such as by intimate data communication or control flow between those
145 | subprograms and other parts of the work.
146 |
147 | The Corresponding Source need not include anything that users
148 | can regenerate automatically from other parts of the Corresponding
149 | Source.
150 |
151 | The Corresponding Source for a work in source code form is that
152 | same work.
153 |
154 | 2. Basic Permissions.
155 |
156 | All rights granted under this License are granted for the term of
157 | copyright on the Program, and are irrevocable provided the stated
158 | conditions are met. This License explicitly affirms your unlimited
159 | permission to run the unmodified Program. The output from running a
160 | covered work is covered by this License only if the output, given its
161 | content, constitutes a covered work. This License acknowledges your
162 | rights of fair use or other equivalent, as provided by copyright law.
163 |
164 | You may make, run and propagate covered works that you do not
165 | convey, without conditions so long as your license otherwise remains
166 | in force. You may convey covered works to others for the sole purpose
167 | of having them make modifications exclusively for you, or provide you
168 | with facilities for running those works, provided that you comply with
169 | the terms of this License in conveying all material for which you do
170 | not control copyright. Those thus making or running the covered works
171 | for you must do so exclusively on your behalf, under your direction
172 | and control, on terms that prohibit them from making any copies of
173 | your copyrighted material outside their relationship with you.
174 |
175 | Conveying under any other circumstances is permitted solely under
176 | the conditions stated below. Sublicensing is not allowed; section 10
177 | makes it unnecessary.
178 |
179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
180 |
181 | No covered work shall be deemed part of an effective technological
182 | measure under any applicable law fulfilling obligations under article
183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or
184 | similar laws prohibiting or restricting circumvention of such
185 | measures.
186 |
187 | When you convey a covered work, you waive any legal power to forbid
188 | circumvention of technological measures to the extent such circumvention
189 | is effected by exercising rights under this License with respect to
190 | the covered work, and you disclaim any intention to limit operation or
191 | modification of the work as a means of enforcing, against the work's
192 | users, your or third parties' legal rights to forbid circumvention of
193 | technological measures.
194 |
195 | 4. Conveying Verbatim Copies.
196 |
197 | You may convey verbatim copies of the Program's source code as you
198 | receive it, in any medium, provided that you conspicuously and
199 | appropriately publish on each copy an appropriate copyright notice;
200 | keep intact all notices stating that this License and any
201 | non-permissive terms added in accord with section 7 apply to the code;
202 | keep intact all notices of the absence of any warranty; and give all
203 | recipients a copy of this License along with the Program.
204 |
205 | You may charge any price or no price for each copy that you convey,
206 | and you may offer support or warranty protection for a fee.
207 |
208 | 5. Conveying Modified Source Versions.
209 |
210 | You may convey a work based on the Program, or the modifications to
211 | produce it from the Program, in the form of source code under the
212 | terms of section 4, provided that you also meet all of these conditions:
213 |
214 | a) The work must carry prominent notices stating that you modified
215 | it, and giving a relevant date.
216 |
217 | b) The work must carry prominent notices stating that it is
218 | released under this License and any conditions added under section
219 | 7. This requirement modifies the requirement in section 4 to
220 | "keep intact all notices".
221 |
222 | c) You must license the entire work, as a whole, under this
223 | License to anyone who comes into possession of a copy. This
224 | License will therefore apply, along with any applicable section 7
225 | additional terms, to the whole of the work, and all its parts,
226 | regardless of how they are packaged. This License gives no
227 | permission to license the work in any other way, but it does not
228 | invalidate such permission if you have separately received it.
229 |
230 | d) If the work has interactive user interfaces, each must display
231 | Appropriate Legal Notices; however, if the Program has interactive
232 | interfaces that do not display Appropriate Legal Notices, your
233 | work need not make them do so.
234 |
235 | A compilation of a covered work with other separate and independent
236 | works, which are not by their nature extensions of the covered work,
237 | and which are not combined with it such as to form a larger program,
238 | in or on a volume of a storage or distribution medium, is called an
239 | "aggregate" if the compilation and its resulting copyright are not
240 | used to limit the access or legal rights of the compilation's users
241 | beyond what the individual works permit. Inclusion of a covered work
242 | in an aggregate does not cause this License to apply to the other
243 | parts of the aggregate.
244 |
245 | 6. Conveying Non-Source Forms.
246 |
247 | You may convey a covered work in object code form under the terms
248 | of sections 4 and 5, provided that you also convey the
249 | machine-readable Corresponding Source under the terms of this License,
250 | in one of these ways:
251 |
252 | a) Convey the object code in, or embodied in, a physical product
253 | (including a physical distribution medium), accompanied by the
254 | Corresponding Source fixed on a durable physical medium
255 | customarily used for software interchange.
256 |
257 | b) Convey the object code in, or embodied in, a physical product
258 | (including a physical distribution medium), accompanied by a
259 | written offer, valid for at least three years and valid for as
260 | long as you offer spare parts or customer support for that product
261 | model, to give anyone who possesses the object code either (1) a
262 | copy of the Corresponding Source for all the software in the
263 | product that is covered by this License, on a durable physical
264 | medium customarily used for software interchange, for a price no
265 | more than your reasonable cost of physically performing this
266 | conveying of source, or (2) access to copy the
267 | Corresponding Source from a network server at no charge.
268 |
269 | c) Convey individual copies of the object code with a copy of the
270 | written offer to provide the Corresponding Source. This
271 | alternative is allowed only occasionally and noncommercially, and
272 | only if you received the object code with such an offer, in accord
273 | with subsection 6b.
274 |
275 | d) Convey the object code by offering access from a designated
276 | place (gratis or for a charge), and offer equivalent access to the
277 | Corresponding Source in the same way through the same place at no
278 | further charge. You need not require recipients to copy the
279 | Corresponding Source along with the object code. If the place to
280 | copy the object code is a network server, the Corresponding Source
281 | may be on a different server (operated by you or a third party)
282 | that supports equivalent copying facilities, provided you maintain
283 | clear directions next to the object code saying where to find the
284 | Corresponding Source. Regardless of what server hosts the
285 | Corresponding Source, you remain obligated to ensure that it is
286 | available for as long as needed to satisfy these requirements.
287 |
288 | e) Convey the object code using peer-to-peer transmission, provided
289 | you inform other peers where the object code and Corresponding
290 | Source of the work are being offered to the general public at no
291 | charge under subsection 6d.
292 |
293 | A separable portion of the object code, whose source code is excluded
294 | from the Corresponding Source as a System Library, need not be
295 | included in conveying the object code work.
296 |
297 | A "User Product" is either (1) a "consumer product", which means any
298 | tangible personal property which is normally used for personal, family,
299 | or household purposes, or (2) anything designed or sold for incorporation
300 | into a dwelling. In determining whether a product is a consumer product,
301 | doubtful cases shall be resolved in favor of coverage. For a particular
302 | product received by a particular user, "normally used" refers to a
303 | typical or common use of that class of product, regardless of the status
304 | of the particular user or of the way in which the particular user
305 | actually uses, or expects or is expected to use, the product. A product
306 | is a consumer product regardless of whether the product has substantial
307 | commercial, industrial or non-consumer uses, unless such uses represent
308 | the only significant mode of use of the product.
309 |
310 | "Installation Information" for a User Product means any methods,
311 | procedures, authorization keys, or other information required to install
312 | and execute modified versions of a covered work in that User Product from
313 | a modified version of its Corresponding Source. The information must
314 | suffice to ensure that the continued functioning of the modified object
315 | code is in no case prevented or interfered with solely because
316 | modification has been made.
317 |
318 | If you convey an object code work under this section in, or with, or
319 | specifically for use in, a User Product, and the conveying occurs as
320 | part of a transaction in which the right of possession and use of the
321 | User Product is transferred to the recipient in perpetuity or for a
322 | fixed term (regardless of how the transaction is characterized), the
323 | Corresponding Source conveyed under this section must be accompanied
324 | by the Installation Information. But this requirement does not apply
325 | if neither you nor any third party retains the ability to install
326 | modified object code on the User Product (for example, the work has
327 | been installed in ROM).
328 |
329 | The requirement to provide Installation Information does not include a
330 | requirement to continue to provide support service, warranty, or updates
331 | for a work that has been modified or installed by the recipient, or for
332 | the User Product in which it has been modified or installed. Access to a
333 | network may be denied when the modification itself materially and
334 | adversely affects the operation of the network or violates the rules and
335 | protocols for communication across the network.
336 |
337 | Corresponding Source conveyed, and Installation Information provided,
338 | in accord with this section must be in a format that is publicly
339 | documented (and with an implementation available to the public in
340 | source code form), and must require no special password or key for
341 | unpacking, reading or copying.
342 |
343 | 7. Additional Terms.
344 |
345 | "Additional permissions" are terms that supplement the terms of this
346 | License by making exceptions from one or more of its conditions.
347 | Additional permissions that are applicable to the entire Program shall
348 | be treated as though they were included in this License, to the extent
349 | that they are valid under applicable law. If additional permissions
350 | apply only to part of the Program, that part may be used separately
351 | under those permissions, but the entire Program remains governed by
352 | this License without regard to the additional permissions.
353 |
354 | When you convey a copy of a covered work, you may at your option
355 | remove any additional permissions from that copy, or from any part of
356 | it. (Additional permissions may be written to require their own
357 | removal in certain cases when you modify the work.) You may place
358 | additional permissions on material, added by you to a covered work,
359 | for which you have or can give appropriate copyright permission.
360 |
361 | Notwithstanding any other provision of this License, for material you
362 | add to a covered work, you may (if authorized by the copyright holders of
363 | that material) supplement the terms of this License with terms:
364 |
365 | a) Disclaiming warranty or limiting liability differently from the
366 | terms of sections 15 and 16 of this License; or
367 |
368 | b) Requiring preservation of specified reasonable legal notices or
369 | author attributions in that material or in the Appropriate Legal
370 | Notices displayed by works containing it; or
371 |
372 | c) Prohibiting misrepresentation of the origin of that material, or
373 | requiring that modified versions of such material be marked in
374 | reasonable ways as different from the original version; or
375 |
376 | d) Limiting the use for publicity purposes of names of licensors or
377 | authors of the material; or
378 |
379 | e) Declining to grant rights under trademark law for use of some
380 | trade names, trademarks, or service marks; or
381 |
382 | f) Requiring indemnification of licensors and authors of that
383 | material by anyone who conveys the material (or modified versions of
384 | it) with contractual assumptions of liability to the recipient, for
385 | any liability that these contractual assumptions directly impose on
386 | those licensors and authors.
387 |
388 | All other non-permissive additional terms are considered "further
389 | restrictions" within the meaning of section 10. If the Program as you
390 | received it, or any part of it, contains a notice stating that it is
391 | governed by this License along with a term that is a further
392 | restriction, you may remove that term. If a license document contains
393 | a further restriction but permits relicensing or conveying under this
394 | License, you may add to a covered work material governed by the terms
395 | of that license document, provided that the further restriction does
396 | not survive such relicensing or conveying.
397 |
398 | If you add terms to a covered work in accord with this section, you
399 | must place, in the relevant source files, a statement of the
400 | additional terms that apply to those files, or a notice indicating
401 | where to find the applicable terms.
402 |
403 | Additional terms, permissive or non-permissive, may be stated in the
404 | form of a separately written license, or stated as exceptions;
405 | the above requirements apply either way.
406 |
407 | 8. Termination.
408 |
409 | You may not propagate or modify a covered work except as expressly
410 | provided under this License. Any attempt otherwise to propagate or
411 | modify it is void, and will automatically terminate your rights under
412 | this License (including any patent licenses granted under the third
413 | paragraph of section 11).
414 |
415 | However, if you cease all violation of this License, then your
416 | license from a particular copyright holder is reinstated (a)
417 | provisionally, unless and until the copyright holder explicitly and
418 | finally terminates your license, and (b) permanently, if the copyright
419 | holder fails to notify you of the violation by some reasonable means
420 | prior to 60 days after the cessation.
421 |
422 | Moreover, your license from a particular copyright holder is
423 | reinstated permanently if the copyright holder notifies you of the
424 | violation by some reasonable means, this is the first time you have
425 | received notice of violation of this License (for any work) from that
426 | copyright holder, and you cure the violation prior to 30 days after
427 | your receipt of the notice.
428 |
429 | Termination of your rights under this section does not terminate the
430 | licenses of parties who have received copies or rights from you under
431 | this License. If your rights have been terminated and not permanently
432 | reinstated, you do not qualify to receive new licenses for the same
433 | material under section 10.
434 |
435 | 9. Acceptance Not Required for Having Copies.
436 |
437 | You are not required to accept this License in order to receive or
438 | run a copy of the Program. Ancillary propagation of a covered work
439 | occurring solely as a consequence of using peer-to-peer transmission
440 | to receive a copy likewise does not require acceptance. However,
441 | nothing other than this License grants you permission to propagate or
442 | modify any covered work. These actions infringe copyright if you do
443 | not accept this License. Therefore, by modifying or propagating a
444 | covered work, you indicate your acceptance of this License to do so.
445 |
446 | 10. Automatic Licensing of Downstream Recipients.
447 |
448 | Each time you convey a covered work, the recipient automatically
449 | receives a license from the original licensors, to run, modify and
450 | propagate that work, subject to this License. You are not responsible
451 | for enforcing compliance by third parties with this License.
452 |
453 | An "entity transaction" is a transaction transferring control of an
454 | organization, or substantially all assets of one, or subdividing an
455 | organization, or merging organizations. If propagation of a covered
456 | work results from an entity transaction, each party to that
457 | transaction who receives a copy of the work also receives whatever
458 | licenses to the work the party's predecessor in interest had or could
459 | give under the previous paragraph, plus a right to possession of the
460 | Corresponding Source of the work from the predecessor in interest, if
461 | the predecessor has it or can get it with reasonable efforts.
462 |
463 | You may not impose any further restrictions on the exercise of the
464 | rights granted or affirmed under this License. For example, you may
465 | not impose a license fee, royalty, or other charge for exercise of
466 | rights granted under this License, and you may not initiate litigation
467 | (including a cross-claim or counterclaim in a lawsuit) alleging that
468 | any patent claim is infringed by making, using, selling, offering for
469 | sale, or importing the Program or any portion of it.
470 |
471 | 11. Patents.
472 |
473 | A "contributor" is a copyright holder who authorizes use under this
474 | License of the Program or a work on which the Program is based. The
475 | work thus licensed is called the contributor's "contributor version".
476 |
477 | A contributor's "essential patent claims" are all patent claims
478 | owned or controlled by the contributor, whether already acquired or
479 | hereafter acquired, that would be infringed by some manner, permitted
480 | by this License, of making, using, or selling its contributor version,
481 | but do not include claims that would be infringed only as a
482 | consequence of further modification of the contributor version. For
483 | purposes of this definition, "control" includes the right to grant
484 | patent sublicenses in a manner consistent with the requirements of
485 | this License.
486 |
487 | Each contributor grants you a non-exclusive, worldwide, royalty-free
488 | patent license under the contributor's essential patent claims, to
489 | make, use, sell, offer for sale, import and otherwise run, modify and
490 | propagate the contents of its contributor version.
491 |
492 | In the following three paragraphs, a "patent license" is any express
493 | agreement or commitment, however denominated, not to enforce a patent
494 | (such as an express permission to practice a patent or covenant not to
495 | sue for patent infringement). To "grant" such a patent license to a
496 | party means to make such an agreement or commitment not to enforce a
497 | patent against the party.
498 |
499 | If you convey a covered work, knowingly relying on a patent license,
500 | and the Corresponding Source of the work is not available for anyone
501 | to copy, free of charge and under the terms of this License, through a
502 | publicly available network server or other readily accessible means,
503 | then you must either (1) cause the Corresponding Source to be so
504 | available, or (2) arrange to deprive yourself of the benefit of the
505 | patent license for this particular work, or (3) arrange, in a manner
506 | consistent with the requirements of this License, to extend the patent
507 | license to downstream recipients. "Knowingly relying" means you have
508 | actual knowledge that, but for the patent license, your conveying the
509 | covered work in a country, or your recipient's use of the covered work
510 | in a country, would infringe one or more identifiable patents in that
511 | country that you have reason to believe are valid.
512 |
513 | If, pursuant to or in connection with a single transaction or
514 | arrangement, you convey, or propagate by procuring conveyance of, a
515 | covered work, and grant a patent license to some of the parties
516 | receiving the covered work authorizing them to use, propagate, modify
517 | or convey a specific copy of the covered work, then the patent license
518 | you grant is automatically extended to all recipients of the covered
519 | work and works based on it.
520 |
521 | A patent license is "discriminatory" if it does not include within
522 | the scope of its coverage, prohibits the exercise of, or is
523 | conditioned on the non-exercise of one or more of the rights that are
524 | specifically granted under this License. You may not convey a covered
525 | work if you are a party to an arrangement with a third party that is
526 | in the business of distributing software, under which you make payment
527 | to the third party based on the extent of your activity of conveying
528 | the work, and under which the third party grants, to any of the
529 | parties who would receive the covered work from you, a discriminatory
530 | patent license (a) in connection with copies of the covered work
531 | conveyed by you (or copies made from those copies), or (b) primarily
532 | for and in connection with specific products or compilations that
533 | contain the covered work, unless you entered into that arrangement,
534 | or that patent license was granted, prior to 28 March 2007.
535 |
536 | Nothing in this License shall be construed as excluding or limiting
537 | any implied license or other defenses to infringement that may
538 | otherwise be available to you under applicable patent law.
539 |
540 | 12. No Surrender of Others' Freedom.
541 |
542 | If conditions are imposed on you (whether by court order, agreement or
543 | otherwise) that contradict the conditions of this License, they do not
544 | excuse you from the conditions of this License. If you cannot convey a
545 | covered work so as to satisfy simultaneously your obligations under this
546 | License and any other pertinent obligations, then as a consequence you may
547 | not convey it at all. For example, if you agree to terms that obligate you
548 | to collect a royalty for further conveying from those to whom you convey
549 | the Program, the only way you could satisfy both those terms and this
550 | License would be to refrain entirely from conveying the Program.
551 |
552 | 13. Use with the GNU Affero General Public License.
553 |
554 | Notwithstanding any other provision of this License, you have
555 | permission to link or combine any covered work with a work licensed
556 | under version 3 of the GNU Affero General Public License into a single
557 | combined work, and to convey the resulting work. The terms of this
558 | License will continue to apply to the part which is the covered work,
559 | but the special requirements of the GNU Affero General Public License,
560 | section 13, concerning interaction through a network will apply to the
561 | combination as such.
562 |
563 | 14. Revised Versions of this License.
564 |
565 | The Free Software Foundation may publish revised and/or new versions of
566 | the GNU General Public License from time to time. Such new versions will
567 | be similar in spirit to the present version, but may differ in detail to
568 | address new problems or concerns.
569 |
570 | Each version is given a distinguishing version number. If the
571 | Program specifies that a certain numbered version of the GNU General
572 | Public License "or any later version" applies to it, you have the
573 | option of following the terms and conditions either of that numbered
574 | version or of any later version published by the Free Software
575 | Foundation. If the Program does not specify a version number of the
576 | GNU General Public License, you may choose any version ever published
577 | by the Free Software Foundation.
578 |
579 | If the Program specifies that a proxy can decide which future
580 | versions of the GNU General Public License can be used, that proxy's
581 | public statement of acceptance of a version permanently authorizes you
582 | to choose that version for the Program.
583 |
584 | Later license versions may give you additional or different
585 | permissions. However, no additional obligations are imposed on any
586 | author or copyright holder as a result of your choosing to follow a
587 | later version.
588 |
589 | 15. Disclaimer of Warranty.
590 |
591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
599 |
600 | 16. Limitation of Liability.
601 |
602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
610 | SUCH DAMAGES.
611 |
612 | 17. Interpretation of Sections 15 and 16.
613 |
614 | If the disclaimer of warranty and limitation of liability provided
615 | above cannot be given local legal effect according to their terms,
616 | reviewing courts shall apply local law that most closely approximates
617 | an absolute waiver of all civil liability in connection with the
618 | Program, unless a warranty or assumption of liability accompanies a
619 | copy of the Program in return for a fee.
620 |
621 | END OF TERMS AND CONDITIONS
622 |
623 | How to Apply These Terms to Your New Programs
624 |
625 | If you develop a new program, and you want it to be of the greatest
626 | possible use to the public, the best way to achieve this is to make it
627 | free software which everyone can redistribute and change under these terms.
628 |
629 | To do so, attach the following notices to the program. It is safest
630 | to attach them to the start of each source file to most effectively
631 | state the exclusion of warranty; and each file should have at least
632 | the "copyright" line and a pointer to where the full notice is found.
633 |
634 |
635 | Copyright (C)
636 |
637 | This program is free software: you can redistribute it and/or modify
638 | it under the terms of the GNU General Public License as published by
639 | the Free Software Foundation, either version 3 of the License, or
640 | (at your option) any later version.
641 |
642 | This program is distributed in the hope that it will be useful,
643 | but WITHOUT ANY WARRANTY; without even the implied warranty of
644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
645 | GNU General Public License for more details.
646 |
647 | You should have received a copy of the GNU General Public License
648 | along with this program. If not, see .
649 |
650 | Also add information on how to contact you by electronic and paper mail.
651 |
652 | If the program does terminal interaction, make it output a short
653 | notice like this when it starts in an interactive mode:
654 |
655 | Copyright (C)
656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
657 | This is free software, and you are welcome to redistribute it
658 | under certain conditions; type `show c' for details.
659 |
660 | The hypothetical commands `show w' and `show c' should show the appropriate
661 | parts of the General Public License. Of course, your program's commands
662 | might be different; for a GUI interface, you would use an "about box".
663 |
664 | You should also get your employer (if you work as a programmer) or school,
665 | if any, to sign a "copyright disclaimer" for the program, if necessary.
666 | For more information on this, and how to apply and follow the GNU GPL, see
667 | .
668 |
669 | The GNU General Public License does not permit incorporating your program
670 | into proprietary programs. If your program is a subroutine library, you
671 | may consider it more useful to permit linking proprietary applications with
672 | the library. If this is what you want to do, use the GNU Lesser General
673 | Public License instead of this License. But first, please read
674 | .
675 |
--------------------------------------------------------------------------------
/Model/model01.pth:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Model/model01.pth
--------------------------------------------------------------------------------
/Model/module.py:
--------------------------------------------------------------------------------
1 | import torch
2 | import torch.nn as nn
3 |
4 | class PatchEmbedding(nn.Module):
5 | def __init__(
6 | self,
7 | in_channels: int = 3,
8 | patch_size: int = 16,
9 | embedding_dim: int = 768
10 | ):
11 | super().__init__()
12 | self.patch_size = patch_size
13 | self.patcher = nn.Conv2d(
14 | in_channels = in_channels,
15 | out_channels = embedding_dim,
16 | kernel_size = patch_size,
17 | stride = patch_size,
18 | padding = 0
19 | )
20 | self.flattener = nn.Flatten(start_dim = 2, end_dim = 3)
21 |
22 | def forward(self, x):
23 | image_size = x.shape[-1]
24 | assert image_size % self.patch_size == 0, f"Input image size must be divisible by patch " \
25 | f"size, please re-check the patch size:" \
26 | f"{self.patch_size} and the image size: " \
27 | f"{image_size}."
28 |
29 | return self.flattener(self.patcher(x)).permute(0, 2, 1)
30 |
31 |
32 | class MSA(nn.Module):
33 | def __init__(
34 | self,
35 | embedding_dim: int = 768,
36 | h = 12,
37 | dropout: float = .0
38 | ):
39 | super().__init__()
40 | self.norm_layer = nn.LayerNorm(normalized_shape = embedding_dim)
41 | self.multi_head_attention = nn.MultiheadAttention(
42 | embed_dim = embedding_dim,
43 | num_heads = h,
44 | dropout = dropout,
45 | batch_first = True
46 | )
47 |
48 | def forward(self, x):
49 | x = self.norm_layer(x)
50 | # we don't need the attention weights but just the layer output, so need_weights = Flase
51 | x, _ = self.multi_head_attention(query = x, key = x, value = x, need_weights = False)
52 | return x
53 |
54 |
55 | class MLP(nn.Module):
56 | def __init__(self, embedding_dim: int = 768, mlp_size: int = 3072, dropout: float = .1):
57 | super().__init__()
58 | self.norm = nn.LayerNorm(normalized_shape = embedding_dim)
59 | self.mlp_body = nn.Sequential(
60 | nn.Linear(in_features = embedding_dim, out_features = mlp_size),
61 | nn.GELU(),
62 | nn.Dropout(p = dropout),
63 | nn.Linear(in_features = mlp_size, out_features = embedding_dim),
64 | nn.Dropout(p = dropout)
65 | )
66 |
67 | def forward(self, x):
68 | x = self.norm(x)
69 | x = self.mlp_body(x)
70 | return x
71 |
72 |
73 | class Encoder(nn.Module):
74 | def __init__(
75 | self, embedding_dim: int = 768, h: int = 12, mlp_size: int = 3072, mlp_dropout:
76 | float = .1, msa_dropout: float = .0
77 | ):
78 | super().__init__()
79 | self.msa = MSA(embedding_dim = embedding_dim, h = h, dropout = msa_dropout)
80 | self.mlp = MLP(embedding_dim = embedding_dim, mlp_size = mlp_size, dropout = mlp_dropout)
81 |
82 | def forward(self, x):
83 | x = self.msa(x) + x
84 | x = self.mlp(x) + x
85 | return x
86 |
87 | class StandardVit(nn.Module):
88 | def __init__(self,
89 | img_size: int = 224,
90 | in_channels: int = 3,
91 | patch_size: int = 16,
92 | transformer_layers_num: int = 12,
93 | embedding_dim: int = 768,
94 | mlp_size: int = 3072,
95 | h: int = 12,
96 | msa_dropout: float = .0,
97 | mlp_dropout: float = .1,
98 | embedding_dropout: float = .1,
99 | num_classes: int = 1000):
100 | super().__init__()
101 | assert img_size % patch_size == 0, f"Image size must be divisible by patch size, " \
102 | f"image size: {img_size}, patch size: {patch_size}."
103 |
104 | self.patch_num = (img_size * img_size) // (patch_size ** 2)
105 |
106 | self.class_embedding = nn.Parameter(data = torch.randn(1, 1, embedding_dim),
107 | requires_grad = True)
108 |
109 | self.position_embedding = nn.Parameter(
110 | data = torch.randn(1, self.patch_num + 1, embedding_dim), requires_grad = True)
111 |
112 | self.embedding_dropout = nn.Dropout(p = embedding_dropout)
113 |
114 | self.patch_embedding = PatchEmbedding(in_channels = in_channels, patch_size = patch_size,
115 | embedding_dim = embedding_dim)
116 |
117 | self.transformer_encoder = nn.Sequential(*[Encoder(embedding_dim = embedding_dim, h = h,
118 | mlp_size = mlp_size, mlp_dropout =
119 | mlp_dropout, msa_dropout =
120 | msa_dropout) for _ in range(transformer_layers_num)])
121 |
122 | self.classifier = nn.Sequential(nn.LayerNorm(normalized_shape = embedding_dim),
123 | nn.Linear(in_features = embedding_dim, out_features = num_classes))
124 |
125 |
126 | def forward(self, x):
127 | batch_size = x.shape[0]
128 |
129 | class_token = self.class_embedding.expand(batch_size, -1, -1)
130 |
131 | x = self.patch_embedding(x)
132 |
133 | x = torch.cat((class_token, x), dim = 1)
134 |
135 | x = self.position_embedding + x
136 |
137 | x = self.embedding_dropout(x)
138 |
139 | x = self.transformer_encoder(x)
140 |
141 | x = self.classifier(x[:, 0])
142 |
143 | return x
144 |
145 |
146 | class TinyVGG(nn.Module):
147 | def __init__(self, in_channel: int, hidden_unit: int, classes_num:int) -> None:
148 | self.block1 = nn.Sequential(
149 | nn.Conv2d(in_channels = in_channel, out_channels = hidden_unit, kernel_size = 3,
150 | stride = 1, padding = 0),
151 | nn.ReLU,
152 | nn.Conv2d(in_channels = hidden_unit, out_channels = hidden_unit, kernel_size = 3,
153 | stride = 1, padding = 0),
154 | nn.ReLU,
155 | nn.MaxPool2d(kernel_size = 2, stride = 2)
156 | )
157 |
158 | self.block2 = nn.Sequential(
159 | nn.nn.Conv2d(in_channels = hidden_unit, out_channels = hidden_unit, kernel_size = 3,
160 | stride = 1, padding = 0),
161 | nn.ReLU,
162 | nn.Conv2d(in_channels = hidden_unit, out_channels = hidden_unit, kernel_size = 3,
163 | stride = 1, padding = 0),
164 | nn.ReLU,
165 | nn.MaxPool2d(kernel_size = 2, stride = 2)
166 | )
167 |
168 | self.block3 = nn.Sequential(
169 | nn.Flatten(),
170 | nn.Linear(in_features = hidden_unit * 13 ** 2, out_features = classes_num)
171 | )
172 |
173 | def forward(self, x):
174 | return self.block3(self.block2(self.block1(x)))
175 |
176 |
177 |
178 |
179 |
180 |
181 |
182 |
183 |
184 |
185 |
186 |
187 |
188 |
189 |
190 |
191 |
--------------------------------------------------------------------------------
/PaperReplicate/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/.DS_Store
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/.DS_Store
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38.md:
--------------------------------------------------------------------------------
1 | # Self-Attention and Multi-head Attention Mechanism
2 |
3 | Tags: Recap
4 |
5 | # Why Self-Attention?
6 |
7 | 
8 |
9 | - The concept of “attention” in deep learning [has its roots in the effort to improve Recurrent Neural Networks (RNNs)](https://arxiv.org/abs/1409.0473) for handling longer sequences or sentences.
10 | - Translating a sentence word-by-word does not work effectively.
11 | - To overcome this issue, attention mechanisms were introduced to give access to all sequence elements at each time step. **The key is to be selective and determine which words are most important in a specific context**.
12 | - In 2017, the transformer architecture introduced a standalone self-attention mechanism, eliminating the need for RNNs altogether.
13 |
14 | # What is Self-Attention?
15 |
16 | - We can think of self-attention as a mechanism that **enhances the information content of an input embedding by including information about the input’s context**. In other words, the self-attention mechanism **enables the model to weigh the importance of different elements in an input sequence and dynamically adjust their influence on the output**.
17 | - This is especially important for language processing tasks, where the meaning of a word can change based on its context within a sentence or document.
18 |
19 | # How to Define Self-Attention?
20 |
21 | ## Embedding Input Sentence
22 |
23 | - For simplicity, here our dictionary dc is restricted to the words that occur in the input sentence. In a real-world application, we would consider all words in the training dataset (typical vocabulary sizes range between 30k to 50k).
24 |
25 | ```python
26 | sentence = "Life is short, eat dessert first"
27 |
28 | # Create Dictionary
29 | dict = {s : i for i, s in enumerate(sorted(sentence.replace(",", "").split()))}
30 | # dict: {'Life': 0, 'dessert': 1, 'eat': 2, 'first': 3, 'is': 4, 'short': 5}
31 |
32 | import torch
33 | sentence_idx = torch.tensor([dict[s] for s in sentence.replace(',', '').split()])
34 | # sentence_idx: tensor([0, 4, 5, 2, 1, 3])
35 | ```
36 |
37 | ### Word Embedding
38 |
39 | - Here, we will use a 16-dimensional embedding such that each input word is represented by a 16-dimensional vector.
40 |
41 | ```python
42 | torch.manual_seed(123)
43 | embeder = torch.nn.Embedding(6, 16)
44 | embedded_sentence = embeder(sentence_idx).detach() # [6, 16]
45 | ```
46 |
47 | ## Define Unnormalized Attention Weights
48 |
49 | ### Define Weight Matrices
50 |
51 | - Self-Attention uses three weight matrices, referred to as $W_q, W_k, W_v$, which are adjusted as model parameters during training.
52 | - These matrics serve to project the inputs into query , key, and value components of the sequence.
53 |
54 | $$\text{Query Sequence: } \mathbf{q}^{(\mathrm{i})}=\mathbf{W}_{\mathrm{q}} \mathbf{x}^{(\mathrm{i})} \text { for } \mathrm{i} \in[1, \mathrm{~T}]$$
55 |
56 |
57 | $$\text{Key Sequence: } \mathbf{k}^{(\mathrm{i})}=\mathbf{W}_{\mathrm{k}} \mathbf{x}^{(\mathrm{i})} \text { for } \mathrm{i} \in[1, \mathrm{~T}]$$
58 |
59 | $$\text{Value Sequence: } \mathbf{v}^{(\mathrm{i})}=\mathbf{W}_{\mathrm{v}} \mathbf{x}^{(\mathrm{i})} \text { for } \mathrm{i} \in[1, \mathrm{~T}]$$
60 |
61 | $$\text{i refers to the token index position in the input sequence}$$
62 |
63 |
64 | - Here, both $q^{(i)}, k^{(i)}$ are vectors of dim $d_k$; $v^{(i)}$ is the vector of dim $d_v$.
65 | $W_q,\ W_k$ have shape $d_k \times d$, $W_v$ has shape $d_v \times d$, $d$ is the embedding dim of each word vector $x^{(i)}$
66 | - Since we are computing the dot-product between the query and key vectors, these two vectors have to contain the same number of elements ($d_q=d_k$). However, the number of elements in the value vector $d_v$ which determines the size of the resulting context vector, is arbitrary.
67 |
68 | ```python
69 | torch.manual_seed(123)
70 |
71 | d = embedded_sentence.shape[1]
72 |
73 | d_q, d_k, d_v = 24, 24, 28
74 |
75 | W_query = torch.nn.Parameter(torch.randn(d_q, d)) # [24, 16]
76 | W_key = torch.nn.Parameter(torch.randn(d_k, d)) # [24, 16]
77 | W_value = torch.nn.Parameter(torch.randn(d_v, d)) # [28, 16]
78 |
79 | ```
80 |
81 | ### Compute the unnormalized weights
82 |
83 | We pick the second words $x^{(2)}$ as example
84 |
85 | ```python
86 | x_2 = embedded_sentence[1]
87 | query_2 = W_query @ x_2 # [24]
88 | key_2 = W_key @ x_2 # [24]
89 | value_2 = W_value @ x_2 # [28]
90 |
91 | # compute the remaining key-value for all inputs
92 | keys = (W_key @ embedded_sentence.T).T # [6, 24]
93 | values = (W_value @ embedded_sentence.T).T #[6, 28]
94 |
95 | ```
96 |
97 | Then compute $\omega_{\mathrm{ij}}=\mathbf{q}^{(\mathrm{i})^{\top}} \mathbf{k}^{(\mathrm{j})}$
98 |
99 | ```python
100 | # compute unnormalized attention weight for the query and 5th input element
101 | omega_24 = query_2.dot(keys[4]) # tensor(-98.1709, grad_fn=)
102 | # then we can let the 2nd word asks every other words
103 | omega_2 = query_2 @ keys.T # (tensor([ 83.1533, 95.5014, -100.8583, 63.5880, -98.1709, 9.3997], grad_fn=)
104 |
105 | ```
106 |
107 | ## Computing Attention Score
108 |
109 | 
110 |
111 | - Then we need the normalized attention weights $\alpha$ by applying the softmax function.
112 | - $1/\sqrt d_k$ is used to scale $w$ before normalization, so that we can ensure that the Euclidean length of the weight vectors will be approximately in the same magnitude.
113 | - This helps prevent the attention weights from becoming too small or too large, which could lead to numerical instability or affect the model’s ability to converge during training.
114 |
115 | ```python
116 | import torch.nn.functional as F
117 |
118 | attention_weights_2 = F.softmax(omega_2 / d_k**0.5, dim=0)
119 | # tensor([7.4329e-02, 9.2430e-01, 3.6185e-18, 1.3699e-03, 6.2628e-18, 2.1523e-08],grad_fn=)
120 |
121 | ```
122 |
123 | ## Compute Context Vector
124 |
125 | 
126 |
127 | - $z^{(2)}$ is an attention-weighted version of our original query input $x^{(2)}$
128 | - The context vector represents the second word in the context of the entire sentence and can be used as input to subsequent layers in a neural network or other models.
129 |
130 | ```python
131 | context_vector_2 = attention_weights_2 @ values # [28]
132 | ```
133 |
134 | ## Dimension Graph
135 |
136 | 
137 |
138 | ## Functionalized Version
139 |
140 | ```python
141 | def calc_attn(embedded_sentence: torch.Tensor,
142 | dim_q: int,
143 | dim_k: int,
144 | dim_v: int,
145 | ):
146 | assert dim_k == dim_q, "dim(K) == dim(Q) must be met!"
147 |
148 | hid_dim = embedded_sentence.shape[-1]
149 | W_Q = nn.Parameter(torch.randn(dim_q, hid_dim))
150 | W_K = nn.Parameter(torch.randn(dim_k, hid_dim))
151 | W_V = nn.Parameter(torch.randn(dim_v, hid_dim))
152 |
153 | Q = (W_K @ embedded_sentence.T).T # (T, d_q)
154 | K = (W_Q @ embedded_sentence.T).T # (T, d_k)
155 | V = (W_V @ embedded_sentence.T).T # (T, d_v)
156 |
157 | attn = torch.softmax(1 / dim_k ** .5 * Q @ K.T, dim = 1) @ V # (T, d_v)
158 |
159 | return attn
160 | ```
161 |
162 | # Multi-Head Attention
163 |
164 | ## Single-Head Attention
165 |
166 | In the scaled dot-product attention, the input sequence was transformed using three matrices representing the query, key, and value.
167 |
168 | - These three matrices can be considered as a single attention head in the context of multi-head attention.
169 |
170 | 
171 |
172 | ## Multi-Head Attention
173 |
174 | As its name implies, multi-head attention involves multiple such heads, each consisting of query, key, and value matrices.
175 |
176 | - The final dimension of the context vector should be $(head, d_v)$
177 |
178 | 
179 |
180 | ### Functionalized Version
181 |
182 | ```python
183 | def multi_head_attn(embedded_sentence: torch.Tensor,
184 | head: int,
185 | dim_q: int,
186 | dim_k: int,
187 | dim_v: int):
188 | assert dim_k == dim_q, "dim(K) == dim(Q) must be met!"
189 | hid_dim = embedded_sentence.shape[-1] # dim
190 | stacked_embedded_sentence = embedded_sentence.repeat(head, 1, 1) # (head, T, dim)
191 |
192 | W_Q = nn.Parameter(torch.randn(head, dim_q, hid_dim)) # (h, d_q, dim)
193 | W_K = nn.Parameter(torch.randn(head, dim_k, hid_dim)) # (h, d_k, dim)
194 | W_V = nn.Parameter(torch.randn(head, dim_v, hid_dim)) # (h, d_v, dim)
195 |
196 | Q = (W_Q @ stacked_embedded_sentence.transpose(1, 2)).transpose(1, 2) # (h, T, d_q)
197 | K = (W_K @ stacked_embedded_sentence.transpose(1, 2)).transpose(1, 2) # (h, T, d_k)
198 | V = (W_V @ stacked_embedded_sentence.transpose(1, 2)).transpose(1, 2) # (h, T, d_v)
199 |
200 | mha = torch.softmax(1 / d_k ** .5 * Q @ K.transpose(1, 2), dim = 2) @ V # (h, T, d_v)
201 |
202 | return mha
203 | ```
204 |
205 | ### For Single Token
206 |
207 | ```python
208 | head = 3
209 |
210 | # (3, 24, 16)
211 | multihead_W_query = torch.nn.Parameter(torch.randn(head, d_q, d))
212 | multihead_W_key = torch.nn.Parameter(torch.randn(head, d_k, d))
213 | # (3, 28, 16)
214 | multihead_W_value = torch.nn.Parameter(torch.randn(head, d_v, d))
215 |
216 | # q, k, v for x_2
217 | multihead_query_2 = multihead_W_query @ x_2 # (3, 24)
218 | multihead_key_2 = multihead_W_key @ x_2 # (3, 24)
219 | multihead_value_2 = multihead_W_value @ x_2 # (3, 24)
220 |
221 | # x_2 asks each other words, then we need to calculate the k, v for other tokens
222 | # first, we need to expand the input sequence embeddings to the number of heads
223 | stacked_inputs = embedded_sentence.T.repeat(head, 1, 1) # (3, 16, 6)
224 |
225 | multihead_keys = torch.bmm(multihead_W_key, stacked_inputs)# (3, 24, 6)
226 | multihead_values = torch.bmm(multihead_W_value, stacked_inputs) # (3, 28, 6)
227 |
228 | # let x_2 asks every token -> unnormalized attention score
229 | multihead_query_2.unsqueeze(1)
230 | multihead_attention_unnormalized_score_2 = torch.bmm(multihead_query_2.unsqueeze(dim = 1),
231 | multihead_keys).squeeze() # (3, 6)
232 | # normalized multihead attention score
233 | multihead_attention_normalized_score_2 = F.softmax(multihead_attention_unnormalized_score_2 / d_k
234 | ** 0.5, dim = 1) # (3, 6)
235 | # multihead attention context score
236 | multihead_context_score_2 = torch.bmm(multihead_attention_normalized_score_2.unsqueeze(1),
237 | multihead_values.permute(0, 2, 1)).squeeze() # (3, 28)
238 |
239 | ```
240 |
241 | # Cross Attention
242 |
243 | - In self-attention, we work with the same input sequence. In cross-attention, we mix or combine two *different* input sequences.
244 | - Two input sequence $x_1$ and $x_2$ can have different numbers of elelments. However, their **embedding dimensions must match**
245 |
246 | 
247 |
248 | If we set $x_1$ = $x_2$, it is equal to self-attention
249 |
250 | - **Generated tokens** from Decoder (sen0) **query** the **tokens from Encoder** (sent1)
251 | - Queries usually come from the decoder, and keys and values usually come from the encoder.
252 |
253 | ## Functionalized Version
254 |
255 | ```python
256 | def cross_attn(embedded_sentence_0: torch.Tensor,
257 | embedded_sentence_1: torch.Tensor,
258 | dim_q: int,
259 | dim_k: int,
260 | dim_v: int):
261 | # Generated tokens from the decoder sentence 0 query the tokens from the encoder sentence 1
262 | assert dim_k == dim_q, "dim(K) == dim(Q) must be met!"
263 | assert embedded_sentence_0.shape[-1] == embedded_sentence_1.shape[-1], "Hid_dim must be same between two sentences"
264 |
265 | hid_dim = embedded_sentence_0.shape[-1]
266 |
267 | W_Q = nn.Parameter(torch.randn(dim_q, hid_dim))
268 | W_K = nn.Parameter(torch.randn(dim_k, hid_dim))
269 | W_V = nn.Parameter(torch.randn(dim_v, hid_dim))
270 |
271 | Q = (W_K @ embedded_sentence_0.T).T # (T_0, d_q)
272 | K = (W_Q @ embedded_sentence_1.T).T # (T_1, d_k)
273 | V = (W_V @ embedded_sentence_1.T).T # (T_1, d_v)
274 |
275 | attn = torch.softmax(1 / dim_k ** .5 * Q @ K.T, dim = 1) @ V # (T_0, d_v)
276 |
277 | return attn
278 | ```
279 |
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 1.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 2.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 3.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 3.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 4.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 4.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 5.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 5.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 6.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled 6.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PaperReplicate/Self_Attention_from_Scratch/Self-Attention and Multi-head Attention Mechanism 036331bdfc7649238f86306bb44bed38/Untitled.png
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/self_attention.py:
--------------------------------------------------------------------------------
1 | import torch
2 | from torch import nn
3 |
4 | def calc_attn(embedded_sentence: torch.Tensor,
5 | dim_q: int,
6 | dim_k: int,
7 | dim_v: int,
8 | ):
9 | assert dim_k == dim_q, "dim(K) == dim(Q) must be met!"
10 |
11 | hid_dim = embedded_sentence.shape[-1]
12 | W_Q = nn.Parameter(torch.randn(dim_q, hid_dim))
13 | W_K = nn.Parameter(torch.randn(dim_k, hid_dim))
14 | W_V = nn.Parameter(torch.randn(dim_v, hid_dim))
15 |
16 | Q = (W_K @ embedded_sentence.T).T # (T, d_q)
17 | K = (W_Q @ embedded_sentence.T).T # (T, d_k)
18 | V = (W_V @ embedded_sentence.T).T # (T, d_v)
19 |
20 | attn = torch.softmax(1 / dim_k ** .5 * Q @ K.T, dim = 1) @ V # (T, d_v)
21 |
22 | return attn
23 |
24 | def cross_attn(embedded_sentence_0: torch.Tensor,
25 | embedded_sentence_1: torch.Tensor,
26 | dim_q: int,
27 | dim_k: int,
28 | dim_v: int):
29 | # Generated tokens from the decoder sentence 0 query the tokens from the encoder sentence 1
30 | assert dim_k == dim_q, "dim(K) == dim(Q) must be met!"
31 | assert embedded_sentence_0.shape[-1] == embedded_sentence_1.shape[-1], "Hid_dim must be same between two sentences"
32 |
33 | hid_dim = embedded_sentence_0.shape[-1]
34 |
35 | W_Q = nn.Parameter(torch.randn(dim_q, hid_dim))
36 | W_K = nn.Parameter(torch.randn(dim_k, hid_dim))
37 | W_V = nn.Parameter(torch.randn(dim_v, hid_dim))
38 |
39 | Q = (W_K @ embedded_sentence_0.T).T # (T_0, d_q)
40 | K = (W_Q @ embedded_sentence_1.T).T # (T_1, d_k)
41 | V = (W_V @ embedded_sentence_1.T).T # (T_1, d_v)
42 |
43 | attn = torch.softmax(1 / dim_k ** .5 * Q @ K.T, dim = 1) @ V # (T_0, d_v)
44 |
45 | return attn
46 |
47 |
48 | def multi_head_attn(embedded_sentence: torch.Tensor,
49 | head: int,
50 | dim_q: int,
51 | dim_k: int,
52 | dim_v: int):
53 | assert dim_k == dim_q, "dim(K) == dim(Q) must be met!"
54 | hid_dim = embedded_sentence.shape[-1] # dim
55 | stacked_embedded_sentence = embedded_sentence.repeat(head, 1, 1) # (head, T, dim)
56 |
57 | W_Q = nn.Parameter(torch.randn(head, dim_q, hid_dim)) # (h, d_q, dim)
58 | W_K = nn.Parameter(torch.randn(head, dim_k, hid_dim)) # (h, d_k, dim)
59 | W_V = nn.Parameter(torch.randn(head, dim_v, hid_dim)) # (h, d_v, dim)
60 |
61 | Q = (W_Q @ stacked_embedded_sentence.transpose(1, 2)).transpose(1, 2) # (h, T, d_q)
62 | K = (W_K @ stacked_embedded_sentence.transpose(1, 2)).transpose(1, 2) # (h, T, d_k)
63 | V = (W_V @ stacked_embedded_sentence.transpose(1, 2)).transpose(1, 2) # (h, T, d_v)
64 |
65 | mha = torch.softmax(1 / d_k ** .5 * Q @ K.transpose(1, 2), dim = 2) @ V # (h, T, d_v)
66 |
67 | return mha
68 |
69 |
70 | if __name__ == '__main__':
71 | sentence = "Life is short, eat dessert first"
72 | # Create Dictionary
73 | dict = {s: i for i, s in enumerate(sorted(sentence.replace(",", "").split()))}
74 | # print(dict)
75 |
76 | sentence_index = torch.tensor([dict[s] for s in sentence.replace(",", "").split()])
77 | # print(sentence_index)
78 |
79 | # word embedding
80 | embeder = torch.nn.Embedding(6, 16)
81 | embedded_sentence = embeder(sentence_index).detach()
82 | # print(embedded_sentence.shape)
83 | d_q, d_k, d_v = 24, 24, 30
84 | multi_head_attn(embedded_sentence, 3, d_q, d_k, d_v)
85 |
86 |
87 |
88 |
89 |
90 |
91 |
92 |
93 |
94 |
--------------------------------------------------------------------------------
/PaperReplicate/Self_Attention_from_Scratch/self_attention_mechanism.ipynb:
--------------------------------------------------------------------------------
1 | {
2 | "cells": [
3 | {
4 | "cell_type": "markdown",
5 | "source": [
6 | "# Embedding an Input Sentence\n",
7 | "\n",
8 | "For simplicity, here our dictionary dc is restricted to the words that occur in the input sentence. In a real-world application, we would consider all words in the training dataset (typical vocabulary sizes range between 30k to 50k).\n",
9 | "\n"
10 | ],
11 | "metadata": {
12 | "collapsed": false
13 | }
14 | },
15 | {
16 | "cell_type": "code",
17 | "execution_count": 3,
18 | "outputs": [
19 | {
20 | "data": {
21 | "text/plain": "{'Life': 0, 'dessert': 1, 'eat': 2, 'first': 3, 'is': 4, 'short': 5}"
22 | },
23 | "execution_count": 3,
24 | "metadata": {},
25 | "output_type": "execute_result"
26 | }
27 | ],
28 | "source": [
29 | "sentence = \"Life is short, eat dessert first\"\n",
30 | "\n",
31 | "# Create Dictionary\n",
32 | "dict = {s : i for i, s in enumerate(sorted(sentence.replace(\",\", \"\").split()))}\n",
33 | "\n",
34 | "dict"
35 | ],
36 | "metadata": {
37 | "collapsed": false,
38 | "ExecuteTime": {
39 | "end_time": "2023-06-29T17:57:31.163972Z",
40 | "start_time": "2023-06-29T17:57:31.161212Z"
41 | }
42 | }
43 | },
44 | {
45 | "cell_type": "markdown",
46 | "source": [
47 | "### Assign the Index to Each Word"
48 | ],
49 | "metadata": {
50 | "collapsed": false
51 | }
52 | },
53 | {
54 | "cell_type": "code",
55 | "execution_count": 4,
56 | "outputs": [
57 | {
58 | "data": {
59 | "text/plain": "tensor([0, 4, 5, 2, 1, 3])"
60 | },
61 | "execution_count": 4,
62 | "metadata": {},
63 | "output_type": "execute_result"
64 | }
65 | ],
66 | "source": [
67 | "import torch\n",
68 | "\n",
69 | "sentence_idx = torch.tensor([dict[s] for s in sentence.replace(',', '').split()])\n",
70 | "sentence_idx"
71 | ],
72 | "metadata": {
73 | "collapsed": false,
74 | "ExecuteTime": {
75 | "end_time": "2023-06-29T18:01:34.498889Z",
76 | "start_time": "2023-06-29T18:01:33.690016Z"
77 | }
78 | }
79 | },
80 | {
81 | "cell_type": "markdown",
82 | "source": [
83 | "### Word Embedding\n",
84 | "Here, we will use a 16-dimensional embedding such that each input word is represented by a 16-dimensional vector."
85 | ],
86 | "metadata": {
87 | "collapsed": false
88 | }
89 | },
90 | {
91 | "cell_type": "code",
92 | "execution_count": 6,
93 | "outputs": [
94 | {
95 | "name": "stdout",
96 | "output_type": "stream",
97 | "text": [
98 | "torch.Size([6, 16])\n"
99 | ]
100 | }
101 | ],
102 | "source": [
103 | "torch.manual_seed(123)\n",
104 | "embeder = torch.nn.Embedding(6, 16)\n",
105 | "embedded_sentence = embeder(sentence_idx).detach()\n",
106 | "\n",
107 | "\n",
108 | "print(embedded_sentence.shape)"
109 | ],
110 | "metadata": {
111 | "collapsed": false,
112 | "ExecuteTime": {
113 | "end_time": "2023-06-29T18:04:16.660623Z",
114 | "start_time": "2023-06-29T18:04:16.627372Z"
115 | }
116 | }
117 | },
118 | {
119 | "cell_type": "markdown",
120 | "source": [
121 | "### Define Weight Matrices"
122 | ],
123 | "metadata": {
124 | "collapsed": false
125 | }
126 | },
127 | {
128 | "cell_type": "code",
129 | "execution_count": 12,
130 | "outputs": [
131 | {
132 | "data": {
133 | "text/plain": "(torch.Size([24, 16]), torch.Size([24, 16]), torch.Size([28, 16]))"
134 | },
135 | "execution_count": 12,
136 | "metadata": {},
137 | "output_type": "execute_result"
138 | }
139 | ],
140 | "source": [
141 | "torch.manual_seed(123)\n",
142 | "\n",
143 | "d = embedded_sentence.shape[1]\n",
144 | "\n",
145 | "d_q, d_k, d_v = 24, 24, 28\n",
146 | "\n",
147 | "W_query = torch.nn.Parameter(torch.randn(d_q, d))\n",
148 | "W_key = torch.nn.Parameter(torch.randn(d_k, d))\n",
149 | "W_value = torch.nn.Parameter(torch.randn(d_v, d))\n",
150 | "\n",
151 | "W_query.shape, W_key.shape, W_value.shape"
152 | ],
153 | "metadata": {
154 | "collapsed": false,
155 | "ExecuteTime": {
156 | "end_time": "2023-06-29T18:39:51.904276Z",
157 | "start_time": "2023-06-29T18:39:51.893251Z"
158 | }
159 | }
160 | },
161 | {
162 | "cell_type": "markdown",
163 | "source": [
164 | "### Computing the Unnormalized Attention Weights\n",
165 | "We pick the second words $x^{(2)}$ as example"
166 | ],
167 | "metadata": {
168 | "collapsed": false
169 | }
170 | },
171 | {
172 | "cell_type": "code",
173 | "execution_count": 8,
174 | "outputs": [
175 | {
176 | "data": {
177 | "text/plain": "(torch.Size([24]), torch.Size([24]), torch.Size([28]))"
178 | },
179 | "execution_count": 8,
180 | "metadata": {},
181 | "output_type": "execute_result"
182 | }
183 | ],
184 | "source": [
185 | "x_2 = embedded_sentence[1]\n",
186 | "query_2 = W_query @ x_2\n",
187 | "key_2 = W_key @ x_2\n",
188 | "value_2 = W_value @ x_2\n",
189 | "\n",
190 | "query_2.shape, key_2.shape, value_2.shape"
191 | ],
192 | "metadata": {
193 | "collapsed": false,
194 | "ExecuteTime": {
195 | "end_time": "2023-06-29T18:32:34.857706Z",
196 | "start_time": "2023-06-29T18:32:34.853013Z"
197 | }
198 | }
199 | },
200 | {
201 | "cell_type": "markdown",
202 | "source": [
203 | "We can then generalize this to compute th remaining key, and value elements for all inputs as well, since we will need them in the next step when we compute the unnormalized attention weights ω:"
204 | ],
205 | "metadata": {
206 | "collapsed": false
207 | }
208 | },
209 | {
210 | "cell_type": "code",
211 | "execution_count": 9,
212 | "outputs": [
213 | {
214 | "data": {
215 | "text/plain": "(torch.Size([6, 24]), torch.Size([6, 28]))"
216 | },
217 | "execution_count": 9,
218 | "metadata": {},
219 | "output_type": "execute_result"
220 | }
221 | ],
222 | "source": [
223 | "keys = (W_key @ embedded_sentence.T).T\n",
224 | "values = (W_value @ embedded_sentence.T).T\n",
225 | "\n",
226 | "keys.shape, values.shape"
227 | ],
228 | "metadata": {
229 | "collapsed": false,
230 | "ExecuteTime": {
231 | "end_time": "2023-06-29T18:34:23.237904Z",
232 | "start_time": "2023-06-29T18:34:23.232098Z"
233 | }
234 | }
235 | },
236 | {
237 | "cell_type": "markdown",
238 | "source": [
239 | "We can then generalize this to compute th remaining key, and value elements for all inputs as well, since we will need them in the next step when we compute the unnormalized attention weights ω\n",
240 | "\n",
241 | "As illustrated in the figure above, we compute $w_{ij}$\n",
242 | " as the dot product between the query and key sequences, $ω_{ij}=q^{(i)}^⊤k^{(j)}$\n",
243 | "\n"
244 | ],
245 | "metadata": {
246 | "collapsed": false
247 | }
248 | },
249 | {
250 | "cell_type": "code",
251 | "execution_count": 13,
252 | "outputs": [
253 | {
254 | "data": {
255 | "text/plain": "tensor(-98.1709, grad_fn=)"
256 | },
257 | "execution_count": 13,
258 | "metadata": {},
259 | "output_type": "execute_result"
260 | }
261 | ],
262 | "source": [
263 | "# Compute the unnormalized attention weights for the query and 5th input word\n",
264 | "omega_24 = query_2.dot(keys[4])\n",
265 | "omega_24"
266 | ],
267 | "metadata": {
268 | "collapsed": false,
269 | "ExecuteTime": {
270 | "end_time": "2023-06-29T18:44:07.066502Z",
271 | "start_time": "2023-06-29T18:44:07.049526Z"
272 | }
273 | }
274 | },
275 | {
276 | "cell_type": "code",
277 | "execution_count": 17,
278 | "outputs": [
279 | {
280 | "data": {
281 | "text/plain": "(tensor([ 83.1533, 95.5014, -100.8583, 63.5880, -98.1709, 9.3997],\n grad_fn=),\n torch.Size([6]))"
282 | },
283 | "execution_count": 17,
284 | "metadata": {},
285 | "output_type": "execute_result"
286 | }
287 | ],
288 | "source": [
289 | "# For all tokens\n",
290 | "omega_2 = query_2 @ keys.T\n",
291 | "omega_2, omega_2.shape"
292 | ],
293 | "metadata": {
294 | "collapsed": false,
295 | "ExecuteTime": {
296 | "end_time": "2023-06-29T18:46:18.727666Z",
297 | "start_time": "2023-06-29T18:46:18.712331Z"
298 | }
299 | }
300 | },
301 | {
302 | "cell_type": "code",
303 | "execution_count": 18,
304 | "outputs": [
305 | {
306 | "data": {
307 | "text/plain": "tensor([7.4329e-02, 9.2430e-01, 3.6185e-18, 1.3699e-03, 6.2628e-18, 2.1523e-08],\n grad_fn=)"
308 | },
309 | "execution_count": 18,
310 | "metadata": {},
311 | "output_type": "execute_result"
312 | }
313 | ],
314 | "source": [
315 | "# Calculate the\n",
316 | "import torch.nn.functional as F\n",
317 | "\n",
318 | "attention_weights_2 = F.softmax(omega_2 / d_k ** .5, dim = 0)\n",
319 | "attention_weights_2"
320 | ],
321 | "metadata": {
322 | "collapsed": false,
323 | "ExecuteTime": {
324 | "end_time": "2023-06-29T18:55:35.660747Z",
325 | "start_time": "2023-06-29T18:55:35.655454Z"
326 | }
327 | }
328 | },
329 | {
330 | "cell_type": "code",
331 | "execution_count": 19,
332 | "outputs": [
333 | {
334 | "data": {
335 | "text/plain": "torch.Size([28])"
336 | },
337 | "execution_count": 19,
338 | "metadata": {},
339 | "output_type": "execute_result"
340 | }
341 | ],
342 | "source": [
343 | "context_vector_2 = attention_weights_2 @ values\n",
344 | "\n",
345 | "context_vector_2.shape"
346 | ],
347 | "metadata": {
348 | "collapsed": false,
349 | "ExecuteTime": {
350 | "end_time": "2023-06-29T19:42:13.014621Z",
351 | "start_time": "2023-06-29T19:42:13.008768Z"
352 | }
353 | }
354 | },
355 | {
356 | "cell_type": "code",
357 | "execution_count": 21,
358 | "outputs": [],
359 | "source": [
360 | "head = 3\n",
361 | "\n",
362 | "# (3, 24, 16)\n",
363 | "multihead_W_query = torch.nn.Parameter(torch.randn(head, d_q, d))\n",
364 | "multihead_W_key = torch.nn.Parameter(torch.randn(head, d_k, d))\n",
365 | "# (3, 28, 16)\n",
366 | "multihead_W_value = torch.nn.Parameter(torch.randn(head, d_v, d))"
367 | ],
368 | "metadata": {
369 | "collapsed": false,
370 | "ExecuteTime": {
371 | "end_time": "2023-06-29T19:50:25.733361Z",
372 | "start_time": "2023-06-29T19:50:25.730348Z"
373 | }
374 | }
375 | },
376 | {
377 | "cell_type": "code",
378 | "execution_count": 29,
379 | "outputs": [],
380 | "source": [
381 | "# q, k, v for x_2\n",
382 | "multihead_query_2 = multihead_W_query @ x_2 # (3, 24)\n",
383 | "multihead_key_2 = multihead_W_key @ x_2 # (3, 24)\n",
384 | "multihead_value_2 = multihead_W_value @ x_2 # (3, 24)\n",
385 | "\n",
386 | "\n",
387 | "# x_2 asks each other words, then we need to calculate the k, v for other tokens\n",
388 | "# first, we need to expand the input sequence embeddings to the number of heads\n",
389 | "stacked_inputs = embedded_sentence.T.repeat(head, 1, 1) # (3, 16, 6)\n",
390 | "\n",
391 | "\n",
392 | "multihead_keys = torch.bmm(multihead_W_key, stacked_inputs)# (3, 24, 6)\n",
393 | "multihead_values = torch.bmm(multihead_W_value, stacked_inputs) # (3, 28, 6)\n",
394 | "\n"
395 | ],
396 | "metadata": {
397 | "collapsed": false,
398 | "ExecuteTime": {
399 | "end_time": "2023-06-29T20:08:08.893835Z",
400 | "start_time": "2023-06-29T20:08:08.884179Z"
401 | }
402 | }
403 | },
404 | {
405 | "cell_type": "code",
406 | "execution_count": 33,
407 | "outputs": [
408 | {
409 | "data": {
410 | "text/plain": "torch.Size([3, 6])"
411 | },
412 | "execution_count": 33,
413 | "metadata": {},
414 | "output_type": "execute_result"
415 | }
416 | ],
417 | "source": [
418 | "# let x_2 asks every token -> unnormalized attention score\n",
419 | "multihead_query_2.unsqueeze(1)\n",
420 | "\n",
421 | "multihead_attention_unnormalized_score_2 = torch.bmm(multihead_query_2.unsqueeze(dim = 1),\n",
422 | " multihead_keys).squeeze() # (3, 6)\n",
423 | "\n",
424 | "multihead_attention_normalized_score_2 = F.softmax(multihead_attention_unnormalized_score_2 / d_k\n",
425 | " ** 0.5, dim = 1) # (3, 6)"
426 | ],
427 | "metadata": {
428 | "collapsed": false,
429 | "ExecuteTime": {
430 | "end_time": "2023-06-29T20:11:33.035618Z",
431 | "start_time": "2023-06-29T20:11:33.032119Z"
432 | }
433 | }
434 | },
435 | {
436 | "cell_type": "code",
437 | "execution_count": 39,
438 | "outputs": [
439 | {
440 | "data": {
441 | "text/plain": "torch.Size([3, 28])"
442 | },
443 | "execution_count": 39,
444 | "metadata": {},
445 | "output_type": "execute_result"
446 | }
447 | ],
448 | "source": [
449 | "multihead_context_score_2 = torch.bmm(multihead_attention_normalized_score_2.unsqueeze(1),\n",
450 | " multihead_values.permute(0, 2, 1)).squeeze() # (3, 28)\n",
451 | "multihead_context_score_2.shape"
452 | ],
453 | "metadata": {
454 | "collapsed": false,
455 | "ExecuteTime": {
456 | "end_time": "2023-06-29T20:17:00.100346Z",
457 | "start_time": "2023-06-29T20:17:00.096019Z"
458 | }
459 | }
460 | },
461 | {
462 | "cell_type": "code",
463 | "execution_count": null,
464 | "outputs": [],
465 | "source": [],
466 | "metadata": {
467 | "collapsed": false
468 | }
469 | }
470 | ],
471 | "metadata": {
472 | "kernelspec": {
473 | "display_name": "Python 3",
474 | "language": "python",
475 | "name": "python3"
476 | },
477 | "language_info": {
478 | "codemirror_mode": {
479 | "name": "ipython",
480 | "version": 2
481 | },
482 | "file_extension": ".py",
483 | "mimetype": "text/x-python",
484 | "name": "python",
485 | "nbconvert_exporter": "python",
486 | "pygments_lexer": "ipython2",
487 | "version": "2.7.6"
488 | }
489 | },
490 | "nbformat": 4,
491 | "nbformat_minor": 0
492 | }
493 |
--------------------------------------------------------------------------------
/PaperReplicate/data_setup.py:
--------------------------------------------------------------------------------
1 | """
2 | Contains functionality for creating PyTorch DataLoaders for
3 | image classification data.
4 | """
5 |
6 | from torchvision import datasets, transforms
7 | from torch.utils.data import DataLoader
8 | import os
9 | from pathlib import Path
10 | import requests
11 | import zipfile
12 |
13 |
14 | def create_dataloaders(
15 | train_dir: str,
16 | test_dir: str,
17 | transform: transforms.Compose,
18 | batch_size: int,
19 | num_workers: int = os.cpu_count()
20 | ):
21 | """Creates training and testing DataLoaders.
22 | Warning: Only suits standard Pytorch Image Path
23 |
24 | Takes in a training directory and testing directory path and turns
25 | them into PyTorch Datasets and then into PyTorch DataLoaders.
26 |
27 | Args:
28 | train_dir: Path to training directory.
29 | test_dir: Path to testing directory.
30 | transform: torchvision transforms to perform on training and testing data.
31 | batch_size: Number of samples per batch in each of the DataLoaders.
32 | num_workers: An integer for number of workers per DataLoader.
33 |
34 | Returns:
35 | A tuple of (train_dataloader, test_dataloader, class_names).
36 | Where class_names is a list of the target classes.
37 | Example usage:
38 | train_dataloader, test_dataloader, class_names = \
39 | = create_dataloaders(train_dir=path/to/train_dir,
40 | test_dir=path/to/test_dir,
41 | transform=some_transform,
42 | batch_size=32,
43 | num_workers=4)
44 | """
45 | # Use ImageFolder to create dataset(s)
46 | train_data = datasets.ImageFolder(train_dir, transform = transform)
47 | test_data = datasets.ImageFolder(test_dir, transform = transform)
48 |
49 | # Get class names
50 | class_names = train_data.classes
51 | class_to_idx = train_data.class_to_idx
52 |
53 | # Turn images into data loaders
54 | # we are using pin_memory = True to avoid unnecessary copying of memory between CPU and GPU
55 | # The benifits of this will likely be seen with larger dataset sizes
56 | # However, pin_memory = True doesn't always improve performance.
57 |
58 | train_dataloader = DataLoader(
59 | train_data,
60 | batch_size = batch_size,
61 | shuffle = True,
62 | num_workers = num_workers,
63 | pin_memory = True,
64 | )
65 | test_dataloader = DataLoader(
66 | test_data,
67 | batch_size = batch_size,
68 | shuffle = False,
69 | num_workers = num_workers,
70 | pin_memory = True,
71 | )
72 |
73 | return train_dataloader, test_dataloader, class_names, class_to_idx
74 |
75 |
76 | def download_data(source: str,
77 | destination: str,
78 | remove_source: bool = True) -> Path:
79 | """Downloads a zipped dataset from source and unzips to destination.
80 |
81 | Args:
82 | source (str): A link to a zipped file containing data.
83 | destination (str): A target directory to unzip data to.
84 | remove_source (bool): Whether to remove the source after downloading and extracting.
85 |
86 | Returns:
87 | pathlib.Path to downloaded data.
88 |
89 | Example usage:
90 | download_data(source="https://github.com/mrdbourke/pytorch-deep-learning/raw/main/data/pizza_steak_sushi.zip",
91 | destination="pizza_steak_sushi")
92 | """
93 | # Setup path to data folder
94 | data_path = Path("data/")
95 | image_path = data_path / destination
96 |
97 | # If the image folder doesn't exist, download it and prepare it...
98 | if image_path.is_dir():
99 | print(f"[INFO] {image_path} directory exists, skipping download.")
100 | else:
101 | print(f"[INFO] Did not find {image_path} directory, creating one...")
102 | image_path.mkdir(parents = True, exist_ok = True)
103 |
104 | # Download pizza, steak, sushi data
105 | target_file = Path(source).name
106 | with open(data_path / target_file, "wb") as f:
107 | request = requests.get(source)
108 | print(f"[INFO] Downloading {target_file} from {source}...")
109 | f.write(request.content)
110 |
111 | # Unzip pizza, steak, sushi data
112 | with zipfile.ZipFile(data_path / target_file, "r") as zip_ref:
113 | print(f"[INFO] Unzipping {target_file} data...")
114 | zip_ref.extractall(image_path)
115 |
116 | # Remove .zip file
117 | if remove_source:
118 | os.remove(data_path / target_file)
119 |
120 | return image_path
121 |
--------------------------------------------------------------------------------
/PaperReplicate/helper_functions.py:
--------------------------------------------------------------------------------
1 | """
2 | A series of helper functions used throughout the course.
3 |
4 | If a function gets defined once and could be used over and over, it'll go in here.
5 | """
6 | import torch
7 | import matplotlib.pyplot as plt
8 | import numpy as np
9 |
10 | import zipfile
11 |
12 | from pathlib import Path
13 |
14 | import requests
15 |
16 | # Walk through an image classification directory and find out how many files (images)
17 | # are in each subdirectory.
18 |
19 | import os
20 | os.environ["TORCH_USE_NNPACK"] = "0"
21 |
22 | def walk_through_dir(dir_path):
23 | """
24 | Walks through dir_path returning its contents.
25 | Args:
26 | dir_path (str): target directory
27 |
28 | Returns:
29 | A print out of:
30 | number of subdiretories in dir_path
31 | number of images (files) in each subdirectory
32 | name of each subdirectory
33 | """
34 | for dirpath, dirnames, filenames in os.walk(dir_path):
35 | print(f"There are {len(dirnames)} directories and {len(filenames)} images in '{dirpath}'.")
36 |
37 | def plot_decision_boundary(model: torch.nn.Module, X: torch.Tensor, y: torch.Tensor):
38 | """Plots decision boundaries of model predicting on X in comparison to y.
39 |
40 | Source - https://madewithml.com/courses/foundations/neural-networks/ (with modifications)
41 | """
42 | # Put everything to CPU (works better with NumPy + Matplotlib)
43 | model.to("cpu")
44 | X, y = X.to("cpu"), y.to("cpu")
45 |
46 | # Setup prediction boundaries and grid
47 | x_min, x_max = X[:, 0].min() - 0.1, X[:, 0].max() + 0.1
48 | y_min, y_max = X[:, 1].min() - 0.1, X[:, 1].max() + 0.1
49 | xx, yy = np.meshgrid(np.linspace(x_min, x_max, 101), np.linspace(y_min, y_max, 101))
50 |
51 | # Make features
52 | X_to_pred_on = torch.from_numpy(np.column_stack((xx.ravel(), yy.ravel()))).float()
53 |
54 | # Make predictions
55 | model.eval()
56 | with torch.inference_mode():
57 | y_logits = model(X_to_pred_on)
58 |
59 | # Test for multi-class or binary and adjust logits to prediction labels
60 | if len(torch.unique(y)) > 2:
61 | y_pred = torch.softmax(y_logits, dim=1).argmax(dim=1) # mutli-class
62 | else:
63 | y_pred = torch.round(torch.sigmoid(y_logits)) # binary
64 |
65 | # Reshape preds and plot
66 | y_pred = y_pred.reshape(xx.shape).detach().numpy()
67 | plt.contourf(xx, yy, y_pred, cmap=plt.cm.RdYlBu, alpha=0.7)
68 | plt.scatter(X[:, 0], X[:, 1], c=y, s=40, cmap=plt.cm.RdYlBu)
69 | plt.xlim(xx.min(), xx.max())
70 | plt.ylim(yy.min(), yy.max())
71 |
72 |
73 | # Plot linear data or training and test and predictions (optional)
74 | def plot_predictions(
75 | train_data, train_labels, test_data, test_labels, predictions=None
76 | ):
77 | """
78 | Plots linear training data and test data and compares predictions.
79 | """
80 | plt.figure(figsize=(10, 7))
81 |
82 | # Plot training data in blue
83 | plt.scatter(train_data, train_labels, c="b", s=4, label="Training data")
84 |
85 | # Plot test data in green
86 | plt.scatter(test_data, test_labels, c="g", s=4, label="Testing data")
87 |
88 | if predictions is not None:
89 | # Plot the predictions in red (predictions were made on the test data)
90 | plt.scatter(test_data, predictions, c="r", s=4, label="Predictions")
91 |
92 | # Show the legend
93 | plt.legend(prop={"size": 14})
94 |
95 |
96 | # Calculate accuracy (a classification metric)
97 | def accuracy_fn(y_true, y_pred):
98 | """Calculates accuracy between truth labels and predictions.
99 |
100 | Args:
101 | y_true (torch.Tensor): Truth labels for predictions.
102 | y_pred (torch.Tensor): Predictions to be compared to predictions.
103 |
104 | Returns:
105 | [torch.float]: Accuracy value between y_true and y_pred, e.g. 78.45
106 | """
107 | correct = torch.eq(y_true, y_pred).sum().item()
108 | acc = (correct / len(y_pred)) * 100
109 | return acc
110 |
111 |
112 | def print_train_time(start, end, device=None):
113 | """Prints difference between start and end time.
114 |
115 | Args:
116 | start (float): Start time of computation (preferred in timeit format).
117 | end (float): End time of computation.
118 | device ([type], optional): Device that compute is running on. Defaults to None.
119 |
120 | Returns:
121 | float: time between start and end in seconds (higher is longer).
122 | """
123 | total_time = end - start
124 | print(f"\nTrain time on {device}: {total_time:.3f} seconds")
125 | return total_time
126 |
127 |
128 | # Plot loss curves of a model
129 | def plot_loss_curves(results):
130 | """Plots training curves of a result dictionary.
131 |
132 | Args:
133 | results (dict): dictionary containing list of values, e.g.
134 | {"train_loss": [...],
135 | "train_acc": [...],
136 | "test_loss": [...],
137 | "test_acc": [...]}
138 | """
139 | loss = results["train_loss"]
140 | test_loss = results["test_loss"]
141 |
142 | accuracy = results["train_acc"]
143 | test_accuracy = results["test_acc"]
144 |
145 | epochs = range(len(results["train_loss"]))
146 |
147 | plt.figure(figsize=(15, 7))
148 |
149 | # Plot loss
150 | plt.subplot(1, 2, 1)
151 | plt.plot(epochs, loss, label="train_loss")
152 | plt.plot(epochs, test_loss, label="test_loss")
153 | plt.title("Loss")
154 | plt.xlabel("Epochs")
155 | plt.legend()
156 |
157 | # Plot accuracy
158 | plt.subplot(1, 2, 2)
159 | plt.plot(epochs, accuracy, label="train_accuracy")
160 | plt.plot(epochs, test_accuracy, label="test_accuracy")
161 | plt.title("Accuracy")
162 | plt.xlabel("Epochs")
163 | plt.legend()
164 |
165 |
166 | # Pred and plot image function from notebook 04
167 | # See creation: https://www.learnpytorch.io/04_pytorch_custom_datasets/#113-putting-custom-image-prediction-together-building-a-function
168 | from typing import List
169 | import torchvision
170 |
171 |
172 | def pred_and_plot_image(
173 | model: torch.nn.Module,
174 | image_path: str,
175 | class_names: List[str] = None,
176 | transform=None,
177 | device: torch.device = "cuda" if torch.cuda.is_available() else "cpu",
178 | ):
179 | """Makes a prediction on a target image with a trained model and plots the image.
180 |
181 | Args:
182 | model (torch.nn.Module): trained PyTorch image classification model.
183 | image_path (str): filepath to target image.
184 | class_names (List[str], optional): different class names for target image. Defaults to None.
185 | transform (_type_, optional): transform of target image. Defaults to None.
186 | device (torch.device, optional): target device to compute on. Defaults to "cuda" if torch.cuda.is_available() else "cpu".
187 |
188 | Returns:
189 | Matplotlib plot of target image and model prediction as title.
190 |
191 | Example usage:
192 | pred_and_plot_image(model=model,
193 | image="some_image.jpeg",
194 | class_names=["class_1", "class_2", "class_3"],
195 | transform=torchvision.transforms.ToTensor(),
196 | device=device)
197 | """
198 |
199 | # 1. Load in image and convert the tensor values to float32
200 | target_image = torchvision.io.read_image(str(image_path)).type(torch.float32)
201 |
202 | # 2. Divide the image pixel values by 255 to get them between [0, 1]
203 | target_image = target_image / 255.0
204 |
205 | # 3. Transform if necessary
206 | if transform:
207 | target_image = transform(target_image)
208 |
209 | # 4. Make sure the model is on the target device
210 | model.to(device)
211 |
212 | # 5. Turn on model evaluation mode and inference mode
213 | model.eval()
214 | with torch.inference_mode():
215 | # Add an extra dimension to the image
216 | target_image = target_image.unsqueeze(dim=0)
217 |
218 | # Make a prediction on image with an extra dimension and send it to the target device
219 | target_image_pred = model(target_image.to(device))
220 |
221 | # 6. Convert logits -> prediction probabilities (using torch.softmax() for multi-class classification)
222 | target_image_pred_probs = torch.softmax(target_image_pred, dim=1)
223 |
224 | # 7. Convert prediction probabilities -> prediction labels
225 | target_image_pred_label = torch.argmax(target_image_pred_probs, dim=1)
226 |
227 | # 8. Plot the image alongside the prediction and prediction probability
228 | plt.imshow(
229 | target_image.squeeze().permute(1, 2, 0)
230 | ) # make sure it's the right size for matplotlib
231 | if class_names:
232 | title = f"Pred: {class_names[target_image_pred_label.cpu()]} | Prob: {target_image_pred_probs.max().cpu():.3f}"
233 | else:
234 | title = f"Pred: {target_image_pred_label} | Prob: {target_image_pred_probs.max().cpu():.3f}"
235 | plt.title(title)
236 | plt.axis(False)
237 |
238 | def set_seeds(seed: int=42):
239 | """Sets random sets for torch operations.
240 |
241 | Args:
242 | seed (int, optional): Random seed to set. Defaults to 42.
243 | """
244 | # Set the seed for general torch operations
245 | torch.manual_seed(seed)
246 | # Set the seed for CUDA torch operations (ones that happen on the GPU)
247 | torch.cuda.manual_seed(seed)
248 |
249 |
250 |
--------------------------------------------------------------------------------
/PaperReplicate/predictions.py:
--------------------------------------------------------------------------------
1 | """
2 | Utility functions to make predictions.
3 |
4 | """
5 | import torch
6 | import torchvision
7 | from torchvision import transforms
8 | import matplotlib.pyplot as plt
9 |
10 | from typing import List, Tuple
11 |
12 | from PIL import Image
13 |
14 | # Set device
15 | device = "cuda" if torch.cuda.is_available() else "cpu"
16 | # Predict on a target image with a target model
17 | def pred_and_plot_image(
18 | model: torch.nn.Module,
19 | class_names: List[str],
20 | image_path: str,
21 | image_size: Tuple[int, int] = (224, 224),
22 | transform: torchvision.transforms = None,
23 | device: torch.device = device,
24 | ):
25 | """Predicts on a target image with a target model.
26 |
27 | Args:
28 | model (torch.nn.Module): A trained (or untrained) PyTorch model to predict on an image.
29 | class_names (List[str]): A list of target classes to map predictions to.
30 | image_path (str): Filepath to target image to predict on.
31 | image_size (Tuple[int, int], optional): Size to transform target image to. Defaults to (224, 224).
32 | transform (torchvision.transforms, optional): Transform to perform on image. Defaults to None which uses ImageNet normalization.
33 | device (torch.device, optional): Target device to perform prediction on. Defaults to device.
34 | """
35 |
36 | # Open image
37 | img = Image.open(image_path)
38 |
39 | # Create transformation for image (if one doesn't exist)
40 | if transform is not None:
41 | image_transform = transform
42 | else:
43 | image_transform = transforms.Compose(
44 | [
45 | transforms.Resize(image_size),
46 | transforms.ToTensor(),
47 | transforms.Normalize(
48 | mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
49 | ),
50 | ]
51 | )
52 |
53 | # Make sure the model is on the target device
54 | model.to(device)
55 |
56 | # Turn on model evaluation mode and inference mode
57 | model.eval()
58 | with torch.inference_mode():
59 | # Transform and add an extra dimension to image (model requires samples in [batch_size, color_channels, height, width])
60 | transformed_image = image_transform(img).unsqueeze(dim=0)
61 |
62 | # Make a prediction on image with an extra dimension and send it to the target device
63 | target_image_pred = model(transformed_image.to(device))
64 |
65 | # Convert logits -> prediction probabilities (using torch.softmax() for multi-class classification)
66 | target_image_pred_probs = torch.softmax(target_image_pred, dim=1)
67 |
68 | # Convert prediction probabilities -> prediction labels
69 | target_image_pred_label = torch.argmax(target_image_pred_probs, dim=1)
70 |
71 | # Plot image with predicted label and probability
72 | plt.figure()
73 | plt.imshow(img)
74 | plt.title(
75 | f"Pred: {class_names[target_image_pred_label]} | Prob: {target_image_pred_probs.max():.3f}"
76 | )
77 | plt.axis(False)
78 |
--------------------------------------------------------------------------------
/PaperReplicate/utils.py:
--------------------------------------------------------------------------------
1 | """
2 | Contains various utility functions for PyTorch model training and saving.
3 | """
4 | import torch
5 | from pathlib import Path
6 | from torchinfo import summary
7 | from torch.utils.tensorboard import SummaryWriter
8 |
9 | def create_writer(experiment_name: str, model_name: str, extra: str=None):
10 | from datetime import datetime
11 | import os
12 |
13 | timestamp = datetime.now().strftime("%Y-%m-%d")
14 |
15 | if extra:
16 | log_dir = os.path.join("runs", timestamp, experiment_name, model_name, extra)
17 | else:
18 | log_dir = os.path.join("runs", timestamp, experiment_name, model_name)
19 |
20 | print(f"[INFO] Created SummaryWriter, saving to: {log_dir}")
21 | return SummaryWriter(log_dir)
22 |
23 |
24 | def set_seed(seed: int = 42, device: torch.device = None):
25 | torch.manual_seed(seed)
26 |
27 | if device == "cuda":
28 | torch.cuda.manual_seed()
29 |
30 |
31 |
32 | def save_model(model: torch.nn.Module,
33 | target_dir: str,
34 | model_name: str):
35 | """Saves a PyTorch model to a target directory.
36 |
37 | Args:
38 | model: A target PyTorch model to save.
39 | target_dir: A directory for saving the model to.
40 | model_name: A filename for the saved model. Should include
41 | either ".pth" or ".pt" as the file extension.
42 |
43 | Example usage:
44 | save_model(model=model_0,
45 | target_dir="models",
46 | model_name="05_going_modular_tingvgg_model.pth")
47 | """
48 | # Create target directory
49 | target_dir_path = Path(target_dir)
50 | target_dir_path.mkdir(parents=True,
51 | exist_ok=True)
52 |
53 | # Create model save path
54 | assert model_name.endswith(".pth") or model_name.endswith(".pt"), "model_name should end with '.pt' or '.pth'"
55 | model_save_path = target_dir_path / model_name
56 |
57 | # Save the model state_dict()
58 | print(f"[INFO] Saving model to: {model_save_path}")
59 | torch.save(obj=model.state_dict(),
60 | f=model_save_path)
61 |
62 |
63 | def device_check():
64 | device = (
65 | "cuda"
66 | if torch.cuda.is_available()
67 | else "mps"
68 | if torch.backends.mps.is_available()
69 | else "cpu"
70 | )
71 | print(f"Using {device} device")
72 | return device
73 |
74 |
75 | import os
76 |
77 | def walk_through_dir(dir_path):
78 | for dirpath, dirnames, filenames in os.walk(dir_path):
79 | print(f"There are {len(dirnames)} directories and {len(filenames)} images in '{dirpath}'.")
80 |
81 |
82 | """
83 | Contains functions for training and testing a PyTorch model.
84 | """
85 | import torch
86 |
87 | from tqdm.auto import tqdm
88 | from typing import Dict, List, Tuple
89 |
90 | def train_step(model: torch.nn.Module,
91 | dataloader: torch.utils.data.DataLoader,
92 | loss_fn: torch.nn.Module,
93 | optimizer: torch.optim.Optimizer, scheduler: None,
94 | device: torch.device) -> Tuple[float, float]:
95 | """Trains a PyTorch model for a single epoch.
96 |
97 | Turns a target PyTorch model to training mode and then
98 | runs through all of the required training steps (forward
99 | pass, loss calculation, optimizer step).
100 |
101 | Args:
102 | model: A PyTorch model to be trained.
103 | dataloader: A DataLoader instance for the model to be trained on.
104 | loss_fn: A PyTorch loss function to minimize.
105 | optimizer: A PyTorch optimizer to help minimize the loss function.
106 | device: A target device to compute on (e.g. "cuda" or "cpu").
107 |
108 | Returns:
109 | A tuple of training loss and training accuracy metrics.
110 | In the form (train_loss, train_accuracy). For example:
111 |
112 | (0.1112, 0.8743)
113 | """
114 | # Put model in train mode
115 | model.train()
116 |
117 | # Setup train loss and train accuracy values
118 | train_loss, train_acc = 0, 0
119 |
120 | # Loop through data loader data batches
121 | for batch, (X, y) in enumerate(dataloader):
122 | # Send data to target device
123 | X, y = X.to(device), y.to(device)
124 |
125 | # 1. Forward pass
126 | y_pred = model(X)
127 |
128 | # 2. Calculate and accumulate loss
129 | loss = loss_fn(y_pred, y)
130 | train_loss += loss.item()
131 |
132 | # 3. Optimizer zero grad
133 | optimizer.zero_grad()
134 |
135 | # 4. Loss backward
136 | loss.backward()
137 |
138 | # 5. Optimizer step
139 | optimizer.step()
140 |
141 | if scheduler:
142 | scheduler.step()
143 |
144 | # Calculate and accumulate accuracy metric across all batches
145 | y_pred_class = torch.argmax(torch.softmax(y_pred, dim = 1), dim = 1)
146 | train_acc += (y_pred_class == y).sum().item() / len(y_pred)
147 |
148 | # Adjust metrics to get average loss and accuracy per batch
149 | train_loss = train_loss / len(dataloader)
150 | train_acc = train_acc / len(dataloader)
151 | return train_loss, train_acc
152 |
153 |
154 | def test_step(model: torch.nn.Module,
155 | dataloader: torch.utils.data.DataLoader,
156 | loss_fn: torch.nn.Module,
157 | device: torch.device) -> Tuple[float, float]:
158 | """Tests a PyTorch model for a single epoch.
159 |
160 | Turns a target PyTorch model to "eval" mode and then performs
161 | a forward pass on a testing dataset.
162 |
163 | Args:
164 | model: A PyTorch model to be tested.
165 | dataloader: A DataLoader instance for the model to be tested on.
166 | loss_fn: A PyTorch loss function to calculate loss on the test data.
167 | device: A target device to compute on (e.g. "cuda" or "cpu").
168 |
169 | Returns:
170 | A tuple of testing loss and testing accuracy metrics.
171 | In the form (test_loss, test_accuracy). For example:
172 |
173 | (0.0223, 0.8985)
174 | """
175 | # Put model in eval mode
176 | model.eval()
177 |
178 | # Setup test loss and test accuracy values
179 | test_loss, test_acc = 0, 0
180 |
181 | # Turn on inference context manager
182 | with torch.inference_mode():
183 | # Loop through DataLoader batches
184 | for batch, (X, y) in enumerate(dataloader):
185 | # Send data to target device
186 | X, y = X.to(device), y.to(device)
187 |
188 | # 1. Forward pass
189 | test_pred_logits = model(X)
190 |
191 | # 2. Calculate and accumulate loss
192 | loss = loss_fn(test_pred_logits, y)
193 | test_loss += loss.item()
194 |
195 | # Calculate and accumulate accuracy
196 | test_pred_labels = test_pred_logits.argmax(dim = 1)
197 | test_acc += ((test_pred_labels == y).sum().item() / len(test_pred_labels))
198 |
199 | # Adjust metrics to get average loss and accuracy per batch
200 | test_loss = test_loss / len(dataloader)
201 | test_acc = test_acc / len(dataloader)
202 | return test_loss, test_acc
203 |
204 |
205 | # Add writer parameter to train()
206 | def train(model: torch.nn.Module,
207 | train_dataloader: torch.utils.data.DataLoader,
208 | test_dataloader: torch.utils.data.DataLoader,
209 | optimizer: torch.optim.Optimizer,
210 | loss_fn: torch.nn.Module,
211 | epochs: int,
212 | device: torch.device,
213 | writer: torch.utils.tensorboard.writer.SummaryWriter # new parameter to take in a writer
214 | ) -> Dict[str, List]:
215 | """Trains and tests a PyTorch model.
216 |
217 | Passes a target PyTorch models through train_step() and test_step()
218 | functions for a number of epochs, training and testing the model
219 | in the same epoch loop.
220 |
221 | Calculates, prints and stores evaluation metrics throughout.
222 |
223 | Stores metrics to specified writer log_dir if present.
224 |
225 | Args:
226 | model: A PyTorch model to be trained and tested.
227 | train_dataloader: A DataLoader instance for the model to be trained on.
228 | test_dataloader: A DataLoader instance for the model to be tested on.
229 | optimizer: A PyTorch optimizer to help minimize the loss function.
230 | loss_fn: A PyTorch loss function to calculate loss on both datasets.
231 | epochs: An integer indicating how many epochs to train for.
232 | device: A target device to compute on (e.g. "cuda" or "cpu").
233 | writer: A SummaryWriter() instance to log model results to.
234 |
235 | Returns:
236 | A dictionary of training and testing loss as well as training and
237 | testing accuracy metrics. Each metric has a value in a list for
238 | each epoch.
239 | """
240 | # Create empty results dictionary
241 | results = {"train_loss": [],
242 | "train_acc": [],
243 | "test_loss": [],
244 | "test_acc": []
245 | }
246 |
247 | # Loop through training and testing steps for a number of epochs
248 | for epoch in tqdm(range(epochs)):
249 | train_loss, train_acc = train_step(model=model,
250 | dataloader=train_dataloader,
251 | loss_fn=loss_fn,
252 | optimizer=optimizer,
253 | scheduler = None,
254 | device=device)
255 | test_loss, test_acc = test_step(model=model,
256 | dataloader=test_dataloader,
257 | loss_fn=loss_fn,
258 | device=device)
259 |
260 | # Print out what's happening
261 | print(
262 | f"Epoch: {epoch+1} | "
263 | f"train_loss: {train_loss:.4f} | "
264 | f"train_acc: {train_acc:.4f} | "
265 | f"test_loss: {test_loss:.4f} | "
266 | f"test_acc: {test_acc:.4f}"
267 | )
268 |
269 | # Update results dictionary
270 | results["train_loss"].append(train_loss)
271 | results["train_acc"].append(train_acc)
272 | results["test_loss"].append(test_loss)
273 | results["test_acc"].append(test_acc)
274 |
275 |
276 | ### New: Use the writer parameter to track experiments ###
277 | # See if there's a writer, if so, log to it
278 | if writer:
279 | writer.add_scalar(tag = "Train Loss", scalar_value = train_loss, global_step = epoch)
280 | writer.add_scalar(tag = "Test Loss", scalar_value = test_loss, global_step = epoch)
281 | writer.add_scalar(tag = "Train Acc", scalar_value = train_acc, global_step = epoch)
282 | writer.add_scalar(tag = "Test Acc", scalar_value = test_acc, global_step = epoch)
283 |
284 | writer.add_graph(model = model,
285 | input_to_model = torch.randn(32, 3, 224, 224).to(device))
286 |
287 |
288 | if writer:
289 | writer.close()
290 |
291 | return results
292 |
293 | import matplotlib.pyplot as plt
294 | # Plot loss curves of a model
295 | def plot_loss_curves(results):
296 | """Plots training curves of a result dictionary.
297 |
298 | Args:
299 | results (dict): dictionary containing list of values, e.g.
300 | {"train_loss": [...],
301 | "train_acc": [...],
302 | "test_loss": [...],
303 | "test_acc": [...]}
304 | """
305 | loss = results["train_loss"]
306 | test_loss = results["test_loss"]
307 |
308 | accuracy = results["train_acc"]
309 | test_accuracy = results["test_acc"]
310 |
311 | epochs = range(len(results["train_loss"]))
312 |
313 | plt.figure(figsize=(15, 7))
314 |
315 | # Plot loss
316 | plt.subplot(1, 2, 1)
317 | plt.plot(epochs, loss, label="train_loss")
318 | plt.plot(epochs, test_loss, label="test_loss")
319 | plt.title("Loss")
320 | plt.xlabel("Epochs")
321 | plt.legend()
322 |
323 | # Plot accuracy
324 | plt.subplot(1, 2, 2)
325 | plt.plot(epochs, accuracy, label="train_accuracy")
326 | plt.plot(epochs, test_accuracy, label="test_accuracy")
327 | plt.title("Accuracy")
328 | plt.xlabel("Epochs")
329 | plt.legend()
--------------------------------------------------------------------------------
/PracticePytorch/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/.DS_Store
--------------------------------------------------------------------------------
/PracticePytorch/03CV.ipynb:
--------------------------------------------------------------------------------
1 | {
2 | "cells": [
3 | {
4 | "cell_type": "code",
5 | "execution_count": 1,
6 | "metadata": {
7 | "collapsed": true,
8 | "ExecuteTime": {
9 | "end_time": "2023-06-12T17:19:47.960360Z",
10 | "start_time": "2023-06-12T17:19:46.651253Z"
11 | }
12 | },
13 | "outputs": [],
14 | "source": [
15 | "import torch\n",
16 | "from torch import nn\n",
17 | "import torchvision\n",
18 | "from torchvision import transforms as trans\n",
19 | "from torchvision import datasets\n",
20 | "\n",
21 | "import matplotlib.pyplot as plt\n",
22 | "\n"
23 | ]
24 | },
25 | {
26 | "cell_type": "code",
27 | "execution_count": 2,
28 | "outputs": [
29 | {
30 | "name": "stdout",
31 | "output_type": "stream",
32 | "text": [
33 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz\n",
34 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw/train-images-idx3-ubyte.gz\n"
35 | ]
36 | },
37 | {
38 | "data": {
39 | "text/plain": " 0%| | 0/26421880 [00:00, ?it/s]",
40 | "application/vnd.jupyter.widget-view+json": {
41 | "version_major": 2,
42 | "version_minor": 0,
43 | "model_id": "e57cb94b5fc049cd82ae5e71e49eb8d1"
44 | }
45 | },
46 | "metadata": {},
47 | "output_type": "display_data"
48 | },
49 | {
50 | "name": "stdout",
51 | "output_type": "stream",
52 | "text": [
53 | "Extracting ./data/FashionMNIST/FashionMNIST/raw/train-images-idx3-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw\n",
54 | "\n",
55 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz\n",
56 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw/train-labels-idx1-ubyte.gz\n"
57 | ]
58 | },
59 | {
60 | "data": {
61 | "text/plain": " 0%| | 0/29515 [00:00, ?it/s]",
62 | "application/vnd.jupyter.widget-view+json": {
63 | "version_major": 2,
64 | "version_minor": 0,
65 | "model_id": "46814269d7c94a21875b8206895c129a"
66 | }
67 | },
68 | "metadata": {},
69 | "output_type": "display_data"
70 | },
71 | {
72 | "name": "stdout",
73 | "output_type": "stream",
74 | "text": [
75 | "Extracting ./data/FashionMNIST/FashionMNIST/raw/train-labels-idx1-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw\n",
76 | "\n",
77 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz\n",
78 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw/t10k-images-idx3-ubyte.gz\n"
79 | ]
80 | },
81 | {
82 | "data": {
83 | "text/plain": " 0%| | 0/4422102 [00:00, ?it/s]",
84 | "application/vnd.jupyter.widget-view+json": {
85 | "version_major": 2,
86 | "version_minor": 0,
87 | "model_id": "7c18c370235d4e72be7b3121389cca78"
88 | }
89 | },
90 | "metadata": {},
91 | "output_type": "display_data"
92 | },
93 | {
94 | "name": "stdout",
95 | "output_type": "stream",
96 | "text": [
97 | "Extracting ./data/FashionMNIST/FashionMNIST/raw/t10k-images-idx3-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw\n",
98 | "\n",
99 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz\n",
100 | "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz\n"
101 | ]
102 | },
103 | {
104 | "data": {
105 | "text/plain": " 0%| | 0/5148 [00:00, ?it/s]",
106 | "application/vnd.jupyter.widget-view+json": {
107 | "version_major": 2,
108 | "version_minor": 0,
109 | "model_id": "519535c933b34a199ede6f6203e00cd9"
110 | }
111 | },
112 | "metadata": {},
113 | "output_type": "display_data"
114 | },
115 | {
116 | "name": "stdout",
117 | "output_type": "stream",
118 | "text": [
119 | "Extracting ./data/FashionMNIST/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz to ./data/FashionMNIST/FashionMNIST/raw\n",
120 | "\n"
121 | ]
122 | }
123 | ],
124 | "source": [
125 | "train_set = datasets.FashionMNIST(\n",
126 | " root = './data/FashionMNIST',\n",
127 | " train = True,\n",
128 | " download = True,\n",
129 | " transform = trans.ToTensor(),\n",
130 | " )\n",
131 | "\n",
132 | "test_set = datasets.FashionMNIST(\n",
133 | " root = './data/FashionMNIST',\n",
134 | " train = False,\n",
135 | " download = True,\n",
136 | " transform = trans.ToTensor()\n",
137 | " )\n",
138 | "\n"
139 | ],
140 | "metadata": {
141 | "collapsed": false,
142 | "ExecuteTime": {
143 | "end_time": "2023-06-12T17:22:06.583677Z",
144 | "start_time": "2023-06-12T17:21:48.900758Z"
145 | }
146 | }
147 | },
148 | {
149 | "cell_type": "code",
150 | "execution_count": 7,
151 | "outputs": [
152 | {
153 | "name": "stdout",
154 | "output_type": "stream",
155 | "text": [
156 | "torch.Size([60000, 28, 28])\n"
157 | ]
158 | },
159 | {
160 | "name": "stderr",
161 | "output_type": "stream",
162 | "text": [
163 | "/Users/lipeiran/opt/anaconda3/envs/DeepLearning/lib/python3.8/site-packages/torchvision/datasets/mnist.py:75: UserWarning: train_data has been renamed data\n",
164 | " warnings.warn(\"train_data has been renamed data\")\n",
165 | "/Users/lipeiran/opt/anaconda3/envs/DeepLearning/lib/python3.8/site-packages/torchvision/datasets/mnist.py:65: UserWarning: train_labels has been renamed targets\n",
166 | " warnings.warn(\"train_labels has been renamed targets\")\n"
167 | ]
168 | },
169 | {
170 | "data": {
171 | "text/plain": "",
172 | "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWMAAAF0CAYAAAD2EVjCAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjYuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8o6BhiAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAkIklEQVR4nO3de3CU1cHH8d8GyJKEEATMDTGNClKF0pY7lUu0psQRRbAidmxoq60VmFK0l8BrgbZDLI60tVQ7tS0FqxXbAtpC1bQSsA1pgdIhpY7lEiECAUGyCQE2Bs77R1/2dU2Ec3CXPUm+n5mdYZ/9ZXOefZYfD7t79gSMMUYAgIRKSvQAAACUMQB4gTIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEumoqKCi1YsEB1dXWJHkpMrVixQpdeeqkaGhoi25qamvStb31L+fn5Sk5OVl5enkpKSnTy5Mmon/3zn/+sbt26af/+/Rd72PAMZYyLpqKiQgsXLmxXZXzixAnNnTtX3/jGN5Senh7ZPm3aND3yyCP64he/qHXr1umee+7RkiVLNHXq1Kifv+GGGzR8+HDNnTv3Yg8dvjHARfLII48YSaa6ujrRQ/lAmpqazDvvvGOMMebxxx83Xbt2NceOHYvcvmnTJiPJPProo1E/t2jRIiPJvPzyy1Hbf/vb35pOnTqZffv2xX3s8BdnxrgoFixYoK997WuSpPz8fAUCAQUCAZWXl0uSVq5cqVGjRiktLU3dunXTpz71KW3bti3qPqZPn65u3bpp165duummm9StWzf17dtXDzzwgMLhcFT2iSee0ODBg9WtWzelp6drwIABLc4+//Wvf+nWW2/VJZdcoq5du+qjH/2oli9fHpUpLy9XIBDQU089pQceeEB9+vRRMBjUrl27Ir9n4sSJ6tGjR+Rn/vrXv0qSbrrppqj7uvnmmyVJv/vd76K2T5w4Ud26ddOTTz5p+3CiHaKMcVHcc889mjVrliRp1apV2rRpkzZt2qSPf/zjWrRokaZNm6ZrrrlGzz33nJ566ik1NDRozJgx+ve//x11P++8845uueUW3XDDDXr++ef1+c9/Xt///vf1ve99L5J59tlndf/992vcuHFavXq11qxZo69+9atqbGyMZF5//XWNHj1aO3bs0GOPPaZVq1bpmmuu0fTp07V48eIW4y8pKdG+ffv0k5/8RL///e+VmZmpN998U1VVVSooKIjKNjU1SZKCwWDU9rPXt2/fHrU9OTlZo0eP1tq1a10fVrQniT41R8fR2ssU+/btM507dzazZs2KyjY0NJjs7Gxzxx13RLYVFxcbSea5556Lyt50003m6quvjlyfOXOm6dGjxznHcuedd5pgMNjipYGioiKTmppq6urqjDHGrF+/3kgyY8eObXEfK1euNJJMZWVl1PY1a9YYSeapp56K2v7zn//cSDL9+/dvcV/z5s0zSUlJ5vjx4+ccN9ovzoyRUC+99JKam5v12c9+Vs3NzZFL165dNW7cuMjLGGcFAgFNnDgxattHPvIR7d27N3J9+PDhqqur07Rp0/T888/ryJEjLX7vK6+8ohtuuEF9+/aN2j59+nSdOHFCmzZtito+ZcqUFvdx4MABSVJmZmbU9qKiIl111VX6xje+obKyMtXV1enFF1/U3Llz1alTJyUltfxrl5mZqTNnzqi2traVRwkdAWWMhDp06JAkadiwYerSpUvUZeXKlS2KNDU1VV27do3aFgwGderUqcj1u+++W7/4xS+0d+9eTZkyRZmZmRoxYoTKysoimaNHjyonJ6fFeHJzcyO3v1tr2bMfU3vveJKTk/XHP/5Rl19+uQoLC3XJJZfo9ttv19y5c3XJJZeoT58+Le7r7H2896Nv6Dg6J3oA6Nh69+4tSfrtb3+rvLy8mN3v5z73OX3uc59TY2OjNm7cqPnz5+vmm2/Wf/7zH+Xl5alXr146ePBgi587e7Z7dlxnBQKB9x3722+/3aKsr7rqKm3atEn79+/X22+/rSuvvFKhUEhf+cpXNHbs2Bb39fbbb7f6e9FxUMa4aM6+gfXus79PfepT6ty5s3bv3t3qSwEfVFpamoqKitTU1KRJkyZpx44dysvL0w033KDVq1frwIEDkbNh6b8TOFJTUzVy5Mjz3veAAQMkSbt379a1117baqZPnz6RM+H/+Z//UVpamr7whS+0yO3Zs0e9evVSVlbWhewm2gHKGBfNoEGDJEk//OEPVVxcrC5duujqq6/Wt7/9bc2bN0979uzRhAkTdMkll+jQoUP6+9//rrS0NC1cuNDp99x7771KSUnRJz7xCeXk5Ki2tlalpaXKyMjQsGHDJEnz58/XH/7wBxUUFOhb3/qWevbsqaefflpr167V4sWLlZGRcd7fM2LECKWkpKiyslK33HJL1G2LFy9Wdna2Lr/8ch06dEjPPfec1qxZo6eeeqrVlykqKys1bty4Vs/A0UEk+h1EdCwlJSUmNzfXJCUlGUlm/fr1xpj/fgKhoKDAdO/e3QSDQZOXl2duv/1286c//Snys8XFxSYtLa3Ffc6fP9+8+6m8fPlyU1BQYLKyskxycrLJzc01d9xxh9m+fXvUz1VVVZmJEyeajIwMk5ycbAYPHmyWLVsWlTn7aYrf/OY3re7P3Xffba655poW2xcuXGiuvPJKEwwGTY8ePcyECRPMxo0bW72PXbt2GUnmd7/7Xau3o2MIGMPq0MCF2rJli4YNG6bKykqNGDHigu7joYce0ooVK7R792517sx/Vjsqyhj4gKZOnarGxkb94Q9/cP7Zuro6XXHFFfrRj36kz3zmM3EYHdoKPtoGfECPPvqohg0bFvWtbbaqq6tVUlKiu+66Kw4jQ1vCmTEAeIAzYwDwAGUMAB6gjAHAA959jubMmTM6cOCA0tPT+QA8gDbNGKOGhgbl5ua2+gVR7+ZdGR84cKDFN2kBQFtWU1Ojyy677JyZuL1M8fjjjys/P19du3bVkCFD9Oqrr1r93LvXEQOA9sCm1+JSxitXrtTs2bM1b948bdu2TWPGjFFRUZH27dt33p/lpQkA7Y1Vr8VjjvXw4cPNfffdF7VtwIAB5pvf/OZ5fzYUChlJXLhw4dJuLqFQ6LzdF/Mz46amJm3dulWFhYVR2wsLC1VRURHrXwcA7ULM38A7cuSITp8+3eJ7WbOyslpdUiYcDket7FtfXx/rIQGA9+L2Bt57XyMxxrT6usnZ75k9e+GTFAA6opiXce/evdWpU6cWZ8GHDx9udRWDkpIShUKhyKWmpibWQwIA78W8jJOTkzVkyJCoxR8lqaysTKNHj26RDwaD6t69e9QFADqauEz6mDNnju6++24NHTpUo0aN0k9/+lPt27dP9913Xzx+HQC0eXEp46lTp+ro0aP69re/rYMHD2rgwIFat25dTFf/BYD2xLvvM66vr7daDBIA2opQKHTel2D51jYA8ABlDAAeoIwBwAOUMQB4gDIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwAGUMAB6gjAHAA5QxAHiAMgYAD3RO9ACA8wkEAtbZeC12np6ebp297rrrrLN//OMfL2Q45+XymHXq1Mk629zcfCHDSRiXx8FFPJ5nnBkDgAcoYwDwAGUMAB6gjAHAA5QxAHiAMgYAD1DGAOAByhgAPEAZA4AHKGMA8ADToeG9pCT7c4bTp09bZ6+66irr7D333GOdPXnypHW2sbHROnvq1Cnr7N///nfrbLymOLtMRXY5xi73G699s51CbozRmTNnrLKcGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwANOh4T2X1YtdpkNff/311tlPfvKT1tk333zTOhsMBq2zqamp1tkbb7zROvuzn/3MOnvo0CHrrMsKyi7HzUW3bt2ss7bTliXpxIkTFzKcc4r5mfGCBQsUCASiLtnZ2bH+NQDQrsTlzPjaa6/Vn/70p8h1lzMbAOiI4lLGnTt35mwYABzE5Q28nTt3Kjc3V/n5+brzzju1Z8+eePwaAGg3Yn5mPGLECK1YsUL9+/fXoUOH9N3vflejR4/Wjh071KtXrxb5cDiscDgcuV5fXx/rIQGA92J+ZlxUVKQpU6Zo0KBB+uQnP6m1a9dKkpYvX95qvrS0VBkZGZFL3759Yz0kAPBe3D9nnJaWpkGDBmnnzp2t3l5SUqJQKBS51NTUxHtIAOCduH/OOBwO67XXXtOYMWNavT0YDDp91hIA2qOYnxk/+OCD2rBhg6qrq/W3v/1Nt99+u+rr61VcXBzrXwUA7UbMz4zffPNNTZs2TUeOHNGll16qkSNHqrKyUnl5ebH+VQDQbsS8jJ999tlY3yU6uKamprjc77Bhw6yzH/rQh6yzLpOcXFZFfumll6yzH/vYx6yzixcvts5u2bLFOltVVWWdfe2116yzw4cPt866HOOKigrr7KZNm6xyxhjrT4jxRUEA4AHKGAA8QBkDgAcoYwDwAGUMAB6gjAHAA5QxAHiAMgYAD1DGAOAByhgAPMDq0EiIQCBgnXVZZdhlVeShQ4daZxsaGqyzaWlp1tn+/fvHJbt582br7K5du6yzLqstjxo1yjo7efJk6+w777xjnXV5HO655x7r7LsXxDiX5uZmvfrqq1ZZzowBwAOUMQB4gDIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeCBgXOaaXgT19fXKyMhI9DDwf1ymLceLy1O0srLSOuuy4rMLl8esubnZOhuvVbJPnTplnT1z5ox19h//+Id11mVKtstjNmHCBOvsFVdcYZ3t06ePdVaSQqGQunfvfs4MZ8YA4AHKGAA8QBkDgAcoYwDwAGUMAB6gjAHAA5QxAHiAMgYAD1DGAOAByhgAPMDq0Dgnz2bLn9exY8esszk5OdbZkydPWmeDwaB1tnNn+7+CLiszu0xxTklJsc66TIceM2aMdXb06NHW2aQk+3PIzMxM6+yLL75onY0HzowBwAOUMQB4gDIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeIDp0GhXUlNTrbMu02pdsidOnLDOhkIh6+zRo0etsy4rX7tMeXdZ+drlMXM5bqdPn7bOukzf7tu3r3U2HpzPjDdu3KiJEycqNzdXgUBAa9asibrdGKMFCxYoNzdXKSkpGj9+vHbs2BGr8QJAu+Rcxo2NjRo8eLCWLl3a6u2LFy/WkiVLtHTpUm3evFnZ2dm68cYb1dDQ8IEHCwDtlfPLFEVFRSoqKmr1NmOMfvCDH2jevHmaPHmyJGn58uXKysrSM888oy996UsfbLQA0E7F9A286upq1dbWqrCwMLItGAxq3LhxqqioiOWvAoB2JaZv4NXW1kqSsrKyorZnZWVp7969rf5MOBxWOByOXK+vr4/lkACgTYjLR9ve+46rMeZ934UtLS1VRkZG5JLodzQBIBFiWsbZ2dmS/v8M+azDhw+3OFs+q6SkRKFQKHKpqamJ5ZAAoE2IaRnn5+crOztbZWVlkW1NTU3asGHD+y6rEgwG1b1796gLAHQ0zq8ZHz9+XLt27Ypcr66u1j//+U/17NlTl19+uWbPnq1FixapX79+6tevnxYtWqTU1FTdddddMR04ALQnzmW8ZcsWFRQURK7PmTNHklRcXKxf/vKX+vrXv66TJ0/q/vvv17FjxzRixAi9/PLLSk9Pj92oAaCdCRjPlv+tr69XRkZGooeB/xOv6a8uU1pdVkXetm2bddZl3+K1OvSBAwess4cOHbLOuqy27DLN2mXacnJysnXWZVKYSz+4vAfl8jz7whe+YJU7ffq0tm3bplAodN6XYPmiIADwAGUMAB6gjAHAA5QxAHiAMgYAD1DGAOAByhgAPEAZA4AHKGMA8ABlDAAeYHVonJPLbPlOnTpZZ12mQ0+dOtU6e/ZrXG289dZb1tmUlBTrrMuKxGlpadZZl+/6bmpqss66TN9+5513rLOdO9vXi8vj26tXL+vsj3/8Y+vsRz/6Ueusy77Z4swYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAeYDo1zcpn26TIF18W//vUv62w4HLbOdunSxTobr6nemZmZ1tlTp05ZZ11WfHZ5HLp27WqddZnqfezYMevsm2++aZ296667rLOPPPKIdbaystI6a4szYwDwAGUMAB6gjAHAA5QxAHiAMgYAD1DGAOAByhgAPEAZA4AHKGMA8ABlDAAe6FDToQOBgHXWZfprUpL9v2kuY3BZiddlRWIXzc3NcblfF+vWrbPONjY2WmdPnjxpnU1OTrbOuqyo7bJCtctz0mXassvzzEW8nr8uj8NHPvIR62woFLLOxgNnxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA80OanQ8dr1V4fpgH7YOzYsdbZKVOmWGc/8YlPWGdPnDhhnXVZFdllirPLKtkuzzOXfXN5rgeDQeusy9Rpl6neLvvmwuW4HT9+3Do7efJk6+zvf/9766wt5zPjjRs3auLEicrNzVUgENCaNWuibp8+fboCgUDUZeTIkbEaLwC0S85l3NjYqMGDB2vp0qXvm5kwYYIOHjwYubh80QsAdETOL1MUFRWpqKjonJlgMKjs7OwLHhQAdDRxeQOvvLxcmZmZ6t+/v+69914dPnw4Hr8GANqNmL+BV1RUpE9/+tPKy8tTdXW1HnroIV1//fXaunVrq28qhMNhhcPhyPX6+vpYDwkAvBfzMp46dWrkzwMHDtTQoUOVl5entWvXtvpuZWlpqRYuXBjrYQBAmxL3zxnn5OQoLy9PO3fubPX2kpIShUKhyKWmpibeQwIA78T9c8ZHjx5VTU2NcnJyWr09GAw6fSYSANoj5zI+fvy4du3aFbleXV2tf/7zn+rZs6d69uypBQsWaMqUKcrJydEbb7yhuXPnqnfv3rrttttiOnAAaE+cy3jLli0qKCiIXJ8zZ44kqbi4WE888YSqqqq0YsUK1dXVKScnRwUFBVq5cqXS09NjN2oAaGcCxmV+40VQX1+vjIyMRA/DSc+ePa2zubm51tl+/frF5X5dpn3279/fOvvuT8Wcj8uK2i6rDKekpFhnDxw4YJ3t0qWLddZlum6vXr2ss01NTdbZ1NRU62xFRYV1tlu3btZZl6n0LqtDu6zi7HLcDh06ZJ398Ic/bJ2V/jvm7t27nzPDFwUBgAcoYwDwAGUMAB6gjAHAA5QxAHiAMgYAD1DGAOAByhgAPEAZA4AHKGMA8ECbXx3aZbHT73znO9bZSy+91Drbo0cP66zLysEuqwHX1dVZZ11Wvm5oaLDOukzXDQQC1tmTJ09aZ12m9t5xxx3W2S1btlhnXb6HxWUK+Yc+9CHrrItBgwZZZ132zeXrcF1WknaZ8u4yfTsvL886Gw+cGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwgLfToZOSkqymzD722GPW95mTk2OddZm27JJ1mfbpwmVFYpfxukxFduGyArjLNNWHH37YOuuyb1/+8petsy6rTp86dco6++c//9k6u2fPHuusyyrk8VrN2mUV53itLP7WW29ZZ+OBM2MA8ABlDAAeoIwBwAOUMQB4gDIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHggYY0yiB/Fu9fX1ysjI0Gc+8xmrKb4u0193795tnXVZVdYlGwwGrbMuXKaTukxFdlnh12UasMvq2y7TX7Ozs62zkyZNss527drVOuuyirPLc2fIkCFxybo8vi5TnF3u12U6vwuXVchd/g7Zrkp/5swZ7d+/X6FQSN27dz9nljNjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwAGUMAB7wdnXot956y2p6ost03fT0dOtsOBy2zrqMwWX6q8sU0fNNtXy3t99+2zq7d+9e66zLvrmszOyygnJzc7N1dvXq1dbZqqoq66zLdOiePXtaZ12mItfV1VlnXVZQdnl8z5w5Y511mYrscr8u06Fd/r7179/fKtfc3Kz9+/dbZZ3OjEtLSzVs2DClp6crMzNTkyZN0uuvvx6VMcZowYIFys3NVUpKisaPH68dO3a4/BoA6HCcynjDhg2aMWOGKisrVVZWpubmZhUWFqqxsTGSWbx4sZYsWaKlS5dq8+bNys7O1o033qiGhoaYDx4A2gunlylefPHFqOvLli1TZmamtm7dqrFjx8oYox/84AeaN2+eJk+eLElavny5srKy9Mwzz+hLX/pS7EYOAO3IB3oDLxQKSfr/172qq6tVW1urwsLCSCYYDGrcuHGqqKj4IL8KANq1C34DzxijOXPm6LrrrtPAgQMlSbW1tZKkrKysqGxWVtb7vhEUDoej3iyrr6+/0CEBQJt1wWfGM2fO1Pbt2/XrX/+6xW3vfQfTGPO+72qWlpYqIyMjcunbt++FDgkA2qwLKuNZs2bphRde0Pr163XZZZdFtp9dZeHsGfJZhw8fbnG2fFZJSYlCoVDk4vIxMQBoL5zK2BijmTNnatWqVXrllVeUn58fdXt+fr6ys7NVVlYW2dbU1KQNGzZo9OjRrd5nMBhU9+7doy4A0NE4vWY8Y8YMPfPMM3r++eeVnp4eOQPOyMhQSkqKAoGAZs+erUWLFqlfv37q16+fFi1apNTUVN11111x2QEAaA+cyviJJ56QJI0fPz5q+7JlyzR9+nRJ0te//nWdPHlS999/v44dO6YRI0bo5Zdfdpr9BgAdjberQw8aNEidOnU6b/7JJ5+0vu8jR45YZ9PS0qyzvXr1ss66TFM9fvy4ddZlOmnnzvb/BrtMPU1NTbXOukyddtk3lxWJXZ76PXr0sM6+exLU+bhMCz927Jh11mUVcpe/F/GaOu1yvykpKdZZl9XCXaZOP/3001a5cDispUuXsjo0ALQVlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwAGUMAB6gjAHAA5QxAHjA29WhbVfjXbVqlfV9fv7zn7fOHjhwwDq7Z88e66zLSsfxmjLsMp3UZcVcm+nrZ7msvn369GnrrMsU5xMnTlhnDx48GJcxuOybyzT2eD3P4rVCtQ+rWb/3WyjP5dChQ1Y5l8eLM2MA8ABlDAAeoIwBwAOUMQB4gDIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHvB2deh4KCoqss4++OCD1tnMzEzrrMtKvC5TRF2m1bpMW3aZDu0yXddlDC6r9ro8nV2mkLtkXR4zl/t1eRxcuNyv7TRgVy6PmcuK5S6rQ2/fvt06e8cdd1hnJbE6NAC0FZQxAHiAMgYAD1DGAOAByhgAPEAZA4AHKGMA8ABlDAAeoIwBwAOUMQB4wNvp0IFAwGqapsvUyHgpKCiwzpaWllpnXaZZu0whT0qy/zfYZdqyy3Rol+nbLg4fPmyddXnq79+/3zrr8pw8fvy4ddblWLhweRxcVmZ2WX3b5TlZVlZmnX3ttdessxUVFdZZV0yHBoA2gjIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPeDsdGm4GDBhgne3du7d11mWF6ssuu8w6+8Ybb1hnXabg7t692zoLXCwxnw5dWlqqYcOGKT09XZmZmZo0aZJef/31qMz06dMj3ytx9jJy5Ej30QNAB+JUxhs2bNCMGTNUWVmpsrIyNTc3q7CwUI2NjVG5CRMm6ODBg5HLunXrYjpoAGhv7L9mS9KLL74YdX3ZsmXKzMzU1q1bNXbs2Mj2YDCo7Ozs2IwQADqAD/QGXigUkiT17Nkzant5ebkyMzPVv39/3XvvvU5fawgAHZHTmfG7GWM0Z84cXXfddRo4cGBke1FRkT796U8rLy9P1dXVeuihh3T99ddr69atCgaDLe4nHA4rHA5HrtfX11/okACgzbrgMp45c6a2b9+uv/zlL1Hbp06dGvnzwIEDNXToUOXl5Wnt2rWaPHlyi/spLS3VwoULL3QYANAuXNDLFLNmzdILL7yg9evXn/fjTDk5OcrLy9POnTtbvb2kpEShUChyqampuZAhAUCb5nRmbIzRrFmztHr1apWXlys/P/+8P3P06FHV1NQoJyen1duDwWCrL18AQEfidGY8Y8YM/epXv9Izzzyj9PR01dbWqra2VidPnpT03/W8HnzwQW3atElvvPGGysvLNXHiRPXu3Vu33XZbXHYAANoDpzPjJ554QpI0fvz4qO3Lli3T9OnT1alTJ1VVVWnFihWqq6tTTk6OCgoKtHLlSqWnp8ds0ADQ3jAdGgDijNWhAaCNoIwBwAOUMQB4gDIGAA9QxgDgAcoYADxAGQOAByhjAPAAZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwAGUMAB6gjAHAA5QxAHiAMgYAD3hXxp6tjwoAH5hNr3lXxg0NDYkeAgDElE2vBYxnp6JnzpzRgQMHlJ6erkAgENleX1+vvn37qqam5rxLXrc17FvbxL61TRdz34wxamhoUG5urpKSzn3u2zmuI7kASUlJuuyyy9739u7du7e7J8dZ7FvbxL61TRdr3zIyMqxy3r1MAQAdEWUMAB5oM2UcDAY1f/58BYPBRA8l5ti3tol9a5t83Tfv3sADgI6ozZwZA0B7RhkDgAcoYwDwAGUMAB5oE2X8+OOPKz8/X127dtWQIUP06quvJnpIMbFgwQIFAoGoS3Z2dqKHdUE2btyoiRMnKjc3V4FAQGvWrIm63RijBQsWKDc3VykpKRo/frx27NiRmME6Ot++TZ8+vcVxHDlyZGIG66C0tFTDhg1Tenq6MjMzNWnSJL3++utRmbZ63Gz2zbfj5n0Zr1y5UrNnz9a8efO0bds2jRkzRkVFRdq3b1+ihxYT1157rQ4ePBi5VFVVJXpIF6SxsVGDBw/W0qVLW7198eLFWrJkiZYuXarNmzcrOztbN954Y5v4LpLz7ZskTZgwIeo4rlu37iKO8MJs2LBBM2bMUGVlpcrKytTc3KzCwkI1NjZGMm31uNnsm+TZcTOeGz58uLnvvvuitg0YMMB885vfTNCIYmf+/Plm8ODBiR5GzEkyq1evjlw/c+aMyc7ONg8//HBk26lTp0xGRob5yU9+koARXrj37psxxhQXF5tbb701IeOJpcOHDxtJZsOGDcaY9nXc3rtvxvh33Lw+M25qatLWrVtVWFgYtb2wsFAVFRUJGlVs7dy5U7m5ucrPz9edd96pPXv2JHpIMVddXa3a2tqo4xgMBjVu3Lh2cxzLy8uVmZmp/v37695779Xhw4cTPSRnoVBIktSzZ09J7eu4vXffzvLpuHldxkeOHNHp06eVlZUVtT0rK0u1tbUJGlXsjBgxQitWrNBLL72kJ598UrW1tRo9erSOHj2a6KHF1Nlj1V6PY1FRkZ5++mm98sorevTRR7V582Zdf/31CofDiR6aNWOM5syZo+uuu04DBw6U1H6OW2v7Jvl33Lz71rbWvPurNKX/Prjv3dYWFRUVRf48aNAgjRo1SldeeaWWL1+uOXPmJHBk8dFej+PUqVMjfx44cKCGDh2qvLw8rV27VpMnT07gyOzNnDlT27dv11/+8pcWt7X14/Z+++bbcfP6zLh3797q1KlTi3+FDx8+3OJf6/YgLS1NgwYN0s6dOxM9lJg6+wmRjnIcc3JylJeX12aO46xZs/TCCy9o/fr1UV9f2x6O2/vtW2sSfdy8LuPk5GQNGTJEZWVlUdvLyso0evToBI0qfsLhsF577TXl5OQkeigxlZ+fr+zs7Kjj2NTUpA0bNrTL43j06FHV1NR4fxyNMZo5c6ZWrVqlV155Rfn5+VG3t+Xjdr59a03Cj1sC3zy08uyzz5ouXbqYn//85+bf//63mT17tklLSzNvvPFGoof2gT3wwAOmvLzc7Nmzx1RWVpqbb77ZpKent8l9a2hoMNu2bTPbtm0zksySJUvMtm3bzN69e40xxjz88MMmIyPDrFq1ylRVVZlp06aZnJwcU19fn+CRn9+59q2hocE88MADpqKiwlRXV5v169ebUaNGmT59+ni/b1/+8pdNRkaGKS8vNwcPHoxcTpw4Ecm01eN2vn3z8bh5X8bGGPPjH//Y5OXlmeTkZPPxj3886uMpbdnUqVNNTk6O6dKli8nNzTWTJ082O3bsSPSwLsj69euNpBaX4uJiY8x/PyY1f/58k52dbYLBoBk7dqypqqpK7KAtnWvfTpw4YQoLC82ll15qunTpYi6//HJTXFxs9u3bl+hhn1dr+yTJLFu2LJJpq8ftfPvm43HjKzQBwANev2YMAB0FZQwAHqCMAcADlDEAeIAyBgAPUMYA4AHKGAA8QBkDgAcoYwDwAGUMAB6gjAHAA5QxAHjgfwHZmFMR+pYM9AAAAABJRU5ErkJggg=="
173 | },
174 | "metadata": {},
175 | "output_type": "display_data"
176 | }
177 | ],
178 | "source": [
179 | "# transfer tensor to np\n",
180 | "print(train_set.train_data.size())\n",
181 | "\n",
182 | "plt.figure(figsize = (4, 4))\n",
183 | "img = train_set.train_data[0].numpy()\n",
184 | "plt.title(train_set.train_labels[0])\n",
185 | "plt.imshow(img, cmap='gray')\n",
186 | "plt.show()\n",
187 | "\n"
188 | ],
189 | "metadata": {
190 | "collapsed": false,
191 | "ExecuteTime": {
192 | "end_time": "2023-06-12T17:27:52.398329Z",
193 | "start_time": "2023-06-12T17:27:52.328533Z"
194 | }
195 | }
196 | },
197 | {
198 | "cell_type": "code",
199 | "execution_count": 9,
200 | "outputs": [
201 | {
202 | "data": {
203 | "text/plain": "['T-shirt/top',\n 'Trouser',\n 'Pullover',\n 'Dress',\n 'Coat',\n 'Sandal',\n 'Shirt',\n 'Sneaker',\n 'Bag',\n 'Ankle boot']"
204 | },
205 | "execution_count": 9,
206 | "metadata": {},
207 | "output_type": "execute_result"
208 | }
209 | ],
210 | "source": [
211 | "train_set.data.__len__()\n",
212 | "\n",
213 | "train_set.classes\n"
214 | ],
215 | "metadata": {
216 | "collapsed": false,
217 | "ExecuteTime": {
218 | "end_time": "2023-06-12T17:36:29.090501Z",
219 | "start_time": "2023-06-12T17:36:29.085122Z"
220 | }
221 | }
222 | },
223 | {
224 | "cell_type": "code",
225 | "execution_count": 10,
226 | "outputs": [],
227 | "source": [
228 | "batch_size = 32\n",
229 | "train_loader = torch.utils.data.DataLoader(\n",
230 | " dataset = train_set,\n",
231 | " batch_size = batch_size,\n",
232 | " shuffle = True\n",
233 | " )\n",
234 | "\n",
235 | "test_loader = torch.utils.data.DataLoader(\n",
236 | " dataset = test_set,\n",
237 | " batch_size = batch_size,\n",
238 | " shuffle = False\n",
239 | " )\n",
240 | "\n"
241 | ],
242 | "metadata": {
243 | "collapsed": false,
244 | "ExecuteTime": {
245 | "end_time": "2023-06-12T17:44:30.188280Z",
246 | "start_time": "2023-06-12T17:44:30.176327Z"
247 | }
248 | }
249 | },
250 | {
251 | "cell_type": "code",
252 | "execution_count": 15,
253 | "outputs": [
254 | {
255 | "data": {
256 | "text/plain": "(torch.Size([32, 1, 28, 28]), torch.Size([32]))"
257 | },
258 | "execution_count": 15,
259 | "metadata": {},
260 | "output_type": "execute_result"
261 | }
262 | ],
263 | "source": [
264 | "train_imgs, train_labels = next(iter(train_loader))\n",
265 | "train_imgs.shape, train_labels.shape"
266 | ],
267 | "metadata": {
268 | "collapsed": false,
269 | "ExecuteTime": {
270 | "end_time": "2023-06-12T17:46:35.856399Z",
271 | "start_time": "2023-06-12T17:46:35.848158Z"
272 | }
273 | }
274 | },
275 | {
276 | "cell_type": "code",
277 | "execution_count": null,
278 | "outputs": [],
279 | "source": [
280 | "# TinyVGG\n",
281 | "class TinyVGG(nn.Module):\n",
282 | " def __init__(self, in_channel: int, output_channel: int, hidden_unit: int):\n",
283 | " super(TinyVGG, self).__init__()\n",
284 | " self.block1 = nn.Sequential(nn.Conv2d(in_channel = in_channel, out_channel = hidden_unit,\n",
285 | " kernel_size = 3,\n",
286 | " stride = 1, padding = 1),\n",
287 | " nn.RelU(),\n",
288 | " nn.Conv2d(in_channel = hidden_unit,\n",
289 | " out_channel = hidden_unit,\n",
290 | " kernel_size = 3, stride = 1, padding = 1),\n",
291 | " nn.RelU(),\n",
292 | " nn.MaxPool2d(kernel_size = 2, stride = 2))\n",
293 | "\n",
294 | " self.block2 = nn.Sequential()"
295 | ],
296 | "metadata": {
297 | "collapsed": false
298 | }
299 | },
300 | {
301 | "cell_type": "code",
302 | "execution_count": null,
303 | "outputs": [],
304 | "source": [],
305 | "metadata": {
306 | "collapsed": false
307 | }
308 | }
309 | ],
310 | "metadata": {
311 | "kernelspec": {
312 | "display_name": "Python 3",
313 | "language": "python",
314 | "name": "python3"
315 | },
316 | "language_info": {
317 | "codemirror_mode": {
318 | "name": "ipython",
319 | "version": 2
320 | },
321 | "file_extension": ".py",
322 | "mimetype": "text/x-python",
323 | "name": "python",
324 | "nbconvert_exporter": "python",
325 | "pygments_lexer": "ipython2",
326 | "version": "2.7.6"
327 | }
328 | },
329 | "nbformat": 4,
330 | "nbformat_minor": 0
331 | }
332 |
--------------------------------------------------------------------------------
/PracticePytorch/backbone.py:
--------------------------------------------------------------------------------
1 | import torch
2 | from torch import nn
3 | from torchvision import transforms
4 | import data_setup, engine
5 | from helper_functions import download_data, set_seeds, plot_loss_curves
6 | from utils import device_check
7 | from module import StandardVit
8 | from torchinfo import summary
9 | from torch.optim.lr_scheduler import LambdaLR
10 |
11 |
12 | # CONFIG
13 | IMG_SIZE = HEIGHT = WIDTH = 224 # image info
14 | CHANNELS = 3
15 | PIN_MEMORY = True # avoid unnecessary copies between CPU and GPU
16 | BATCH_SIZE = 512 # 4096 by default
17 | PATCH_SIZE = 16
18 | TRANSFORMER_LAYER_NUM = 12
19 | HIDDEN_UNIT = 3072
20 | HEAD = 12
21 | MSA_DROPOUT = .0
22 | MLP_DROPOUT = .1
23 | EMBEDDING_DROPOUT = .1
24 | BETA_1 = 0.9 # for Adam optimizer
25 | BETA_2 = 0.999
26 | WEIGHT_DECAY = 0.3
27 | LEARNING_RATE = 0.008
28 |
29 | # number of patches
30 | N = (HEIGHT * WIDTH) // (PATCH_SIZE ** 2)
31 |
32 |
33 | # Get data and set image_path
34 | image_path = download_data(
35 | source = "https://github.com/mrdbourke/pytorch-deep-learning/raw/main/data/pizza_steak_sushi.zip",
36 | destination = "pizza_steak_sushi"
37 | )
38 | train_dir = image_path / "train"
39 | test_dir = image_path / "test"
40 |
41 | # Create Dataloader
42 | img_transform = transforms.Compose(
43 | [
44 | transforms.Resize([IMG_SIZE, IMG_SIZE]),
45 | transforms.ToTensor()
46 | ]
47 | )
48 |
49 | train_data, test_data, class_names = data_setup.create_dataloaders(
50 | train_dir = train_dir,
51 | test_dir = test_dir,
52 | transform = img_transform,
53 | batch_size = BATCH_SIZE)
54 |
55 | # Build the model
56 | vit = StandardVit(img_size = IMG_SIZE,
57 | in_channels = CHANNELS,
58 | patch_size = PATCH_SIZE,
59 | transformer_layers_num = TRANSFORMER_LAYER_NUM,
60 | mlp_size = HIDDEN_UNIT,
61 | h = HEAD,
62 | msa_dropout = MSA_DROPOUT,
63 | mlp_dropout = MLP_DROPOUT,
64 | embedding_dropout = EMBEDDING_DROPOUT,
65 | num_classes = len(class_names))
66 |
67 | vit = nn.DataParallel(vit)
68 |
69 | # Print the model info
70 | model_config = summary( model = vit, input_size = (BATCH_SIZE, CHANNELS, HEIGHT, WIDTH),
71 | col_names = ["input_size", "output_size", "num_params", "trainable"],
72 | col_width = 20,
73 | row_settings = ["var_names"])
74 |
75 | # Train
76 | optimizer = torch.optim.Adam(params = vit.parameters(),
77 | lr = LEARNING_RATE,
78 | betas = (BETA_1, BETA_2),
79 | weight_decay = WEIGHT_DECAY)
80 | num_training_steps = 10000
81 | num_warmup_steps = 1000
82 |
83 | # Define a lambda function for the learning rate schedule
84 | lr_lambda = lambda step: step ** (-0.5) if step != 0 else 1e-4
85 | scheduler = LambdaLR(optimizer, lr_lambda)
86 |
87 | loss = nn.CrossEntropyLoss()
88 | set_seeds()
89 |
90 | result = engine.train(vit, train_data, test_data, optimizer, scheduler, loss, epochs = 1000,
91 | device = device_check())
92 |
93 | plot_loss_curves(result)
94 |
--------------------------------------------------------------------------------
/PracticePytorch/faces/0805personali01.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/0805personali01.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/1084239450_e76e00b7e7.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/1084239450_e76e00b7e7.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/10comm-decarlo.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/10comm-decarlo.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/110276240_bec305da91.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/110276240_bec305da91.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/1198_0_861.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/1198_0_861.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/137341995_e7c48e9a75.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/137341995_e7c48e9a75.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/1383023626_8a49e4879a.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/1383023626_8a49e4879a.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/144044282_87cf3ff76e.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/144044282_87cf3ff76e.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/152601997_ec6429a43c.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/152601997_ec6429a43c.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/1549040388_b99e9fa295.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/1549040388_b99e9fa295.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/1878519279_f905d4f34e.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/1878519279_f905d4f34e.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2046713398_91aaa6fe1c.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2046713398_91aaa6fe1c.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2173711035_dbd53b4f9f.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2173711035_dbd53b4f9f.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2210514040_6b03ff2629.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2210514040_6b03ff2629.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2322901504_08122b01ba.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2322901504_08122b01ba.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2327253037_66a61ea6fe.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2327253037_66a61ea6fe.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2328398005_d328a70b4c.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2328398005_d328a70b4c.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2370961440_6bc8ce346c.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2370961440_6bc8ce346c.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2382SJ8.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2382SJ8.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/252418361_440b75751b.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/252418361_440b75751b.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/262007783_943bbcf613.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/262007783_943bbcf613.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2633371780_45b740b670.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2633371780_45b740b670.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2647088981_60e9fe40cd.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2647088981_60e9fe40cd.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2711409561_a0786a3d3d.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2711409561_a0786a3d3d.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2722779845_7fcb64a096.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2722779845_7fcb64a096.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2795838930_0cc5aa5f41.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2795838930_0cc5aa5f41.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2902323565_100017b63c.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2902323565_100017b63c.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2902760364_89c50bde40.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2902760364_89c50bde40.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/2956581526_cd803f2daa.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/2956581526_cd803f2daa.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/297448785_b2dda4b2c0.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/297448785_b2dda4b2c0.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/299733036_fff5ea6f8e.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/299733036_fff5ea6f8e.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/303808204_1f744bc407.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/303808204_1f744bc407.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3074791551_baee7fa0c1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3074791551_baee7fa0c1.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3152653555_68322314f3.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3152653555_68322314f3.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3264867945_fe18d442c1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3264867945_fe18d442c1.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3273658251_b95f65c244.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3273658251_b95f65c244.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3298715079_5af7c78fcb.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3298715079_5af7c78fcb.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3325611505_ddc7beffa1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3325611505_ddc7beffa1.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3362762930_24f76cb89c.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3362762930_24f76cb89c.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/343583208_e986824d77.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/343583208_e986824d77.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3461016494_56cce9c984.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3461016494_56cce9c984.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/348272697_832ce65324.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/348272697_832ce65324.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3534188114_2108895291.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3534188114_2108895291.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3534189272_8ef88ba368.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3534189272_8ef88ba368.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3555944509_7b477069c6.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3555944509_7b477069c6.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3574737496_6ee8207045.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3574737496_6ee8207045.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/362167809_d5a5dcbfdb.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/362167809_d5a5dcbfdb.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/363149951_8be04dc6c0.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/363149951_8be04dc6c0.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3638950581_3387685d3a.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3638950581_3387685d3a.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3646828311_bfeb429ef7.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3646828311_bfeb429ef7.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3689162471_5f9ffb5aa0.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3689162471_5f9ffb5aa0.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3718903026_c1bf5dfcf8.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3718903026_c1bf5dfcf8.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3790616528_297c0ac935.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3790616528_297c0ac935.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3855944735_e252959937.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3855944735_e252959937.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3856149136_d4595ffdd4.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3856149136_d4595ffdd4.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/3872768751_e60d7fdbd5.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/3872768751_e60d7fdbd5.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/529447797_0f9d2fb756.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/529447797_0f9d2fb756.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/57635685_d41c98f8ca.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/57635685_d41c98f8ca.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/809285949_6889026b53.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/809285949_6889026b53.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/92053278_be61a225d2.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/92053278_be61a225d2.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/96063776_bdb3617b64.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/96063776_bdb3617b64.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/97308305_4b737d0873.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/97308305_4b737d0873.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/britney-bald.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/britney-bald.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/create_landmark_dataset.py:
--------------------------------------------------------------------------------
1 | """Create a sample face landmarks dataset.
2 |
3 | Adapted from dlib/python_examples/face_landmark_detection.py
4 | See this file for more explanation.
5 |
6 | Download a trained facial shape predictor from:
7 | http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2
8 | """
9 | import dlib
10 | import glob
11 | import csv
12 | from skimage import io
13 |
14 | detector = dlib.get_frontal_face_detector()
15 | predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
16 | num_landmarks = 68
17 |
18 | with open('face_landmarks.csv', 'w', newline= '') as csvfile:
19 | csv_writer = csv.writer(csvfile)
20 |
21 | header = ['image_name']
22 | for i in range(num_landmarks):
23 | header += ['part_{}_x'.format(i), 'part_{}_y'.format(i)]
24 |
25 | csv_writer.writerow(header)
26 |
27 | for f in glob.glob('*.jpg'):
28 | img = io.imread(f)
29 | dets = detector(img, 1) # face detection
30 |
31 | # ignore all the files with no or more than one faces detected.
32 | if len(dets) == 1:
33 | row = [f]
34 |
35 | d = dets[0]
36 | # Get the landmarks/parts for the face in box d.
37 | shape = predictor(img, d)
38 | for i in range(num_landmarks):
39 | part_i_x = shape.part(i).x
40 | part_i_y = shape.part(i).y
41 | row += [part_i_x, part_i_y]
42 |
43 | csv_writer.writerow(row)
44 |
--------------------------------------------------------------------------------
/PracticePytorch/faces/deeny.peggy.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/deeny.peggy.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/face_landmarks.csv:
--------------------------------------------------------------------------------
1 | image_name,part_0_x,part_0_y,part_1_x,part_1_y,part_2_x,part_2_y,part_3_x,part_3_y,part_4_x,part_4_y,part_5_x,part_5_y,part_6_x,part_6_y,part_7_x,part_7_y,part_8_x,part_8_y,part_9_x,part_9_y,part_10_x,part_10_y,part_11_x,part_11_y,part_12_x,part_12_y,part_13_x,part_13_y,part_14_x,part_14_y,part_15_x,part_15_y,part_16_x,part_16_y,part_17_x,part_17_y,part_18_x,part_18_y,part_19_x,part_19_y,part_20_x,part_20_y,part_21_x,part_21_y,part_22_x,part_22_y,part_23_x,part_23_y,part_24_x,part_24_y,part_25_x,part_25_y,part_26_x,part_26_y,part_27_x,part_27_y,part_28_x,part_28_y,part_29_x,part_29_y,part_30_x,part_30_y,part_31_x,part_31_y,part_32_x,part_32_y,part_33_x,part_33_y,part_34_x,part_34_y,part_35_x,part_35_y,part_36_x,part_36_y,part_37_x,part_37_y,part_38_x,part_38_y,part_39_x,part_39_y,part_40_x,part_40_y,part_41_x,part_41_y,part_42_x,part_42_y,part_43_x,part_43_y,part_44_x,part_44_y,part_45_x,part_45_y,part_46_x,part_46_y,part_47_x,part_47_y,part_48_x,part_48_y,part_49_x,part_49_y,part_50_x,part_50_y,part_51_x,part_51_y,part_52_x,part_52_y,part_53_x,part_53_y,part_54_x,part_54_y,part_55_x,part_55_y,part_56_x,part_56_y,part_57_x,part_57_y,part_58_x,part_58_y,part_59_x,part_59_y,part_60_x,part_60_y,part_61_x,part_61_y,part_62_x,part_62_y,part_63_x,part_63_y,part_64_x,part_64_y,part_65_x,part_65_y,part_66_x,part_66_y,part_67_x,part_67_y
2 | 0805personali01.jpg,27,83,27,98,29,113,33,127,39,139,49,150,60,159,73,166,87,168,100,166,111,160,120,151,128,141,133,128,137,116,138,102,138,89,44,70,53,66,63,65,73,66,82,70,103,72,111,69,120,68,128,69,135,73,92,82,92,91,92,100,92,109,82,117,86,119,91,120,95,119,99,118,55,82,62,78,70,79,76,84,69,86,61,85,106,86,111,81,119,81,125,85,120,88,112,88,71,137,78,131,85,127,90,129,94,129,99,134,103,142,97,144,92,145,88,145,83,144,77,141,75,137,85,134,89,135,93,136,100,141,93,135,89,135,84,134
3 | 1084239450_e76e00b7e7.jpg,70,236,71,257,75,278,82,299,90,320,100,340,111,359,126,375,149,379,175,376,197,364,218,346,236,322,249,296,254,266,256,237,256,207,65,210,69,197,80,193,93,196,106,201,143,196,160,185,179,180,198,182,212,193,127,212,126,230,126,248,125,266,113,279,122,283,132,285,143,281,154,277,81,224,87,216,98,215,110,222,99,225,88,227,159,216,170,207,181,206,192,212,183,215,171,216,110,314,117,310,126,308,135,309,147,307,164,306,184,307,167,317,152,321,139,323,130,322,120,318,114,313,127,312,136,313,148,311,179,308,149,312,137,314,128,312
4 | 10comm-decarlo.jpg,66,114,65,128,67,142,68,156,72,169,80,180,91,189,103,195,115,197,128,197,139,192,148,184,154,173,160,162,164,150,169,138,171,126,80,101,87,95,98,92,108,96,116,101,139,103,147,100,156,100,164,105,168,113,127,113,127,121,126,128,126,136,115,143,119,145,124,147,128,146,132,146,90,112,97,108,104,109,110,114,103,115,96,114,139,118,146,114,153,116,157,121,152,122,146,121,101,163,109,157,117,155,123,157,129,156,134,160,139,168,133,171,126,172,120,172,114,171,107,168,104,162,116,161,122,162,128,162,136,167,127,166,121,165,116,164
5 | 110276240_bec305da91.jpg,42,140,45,161,51,180,61,200,73,220,89,238,108,255,129,268,155,268,180,261,203,244,222,224,238,199,247,172,248,144,245,116,241,87,39,114,45,102,57,97,72,96,87,99,120,91,136,78,155,70,178,66,199,71,108,114,111,132,115,150,118,168,107,180,116,182,127,182,138,177,147,171,57,134,65,127,77,122,92,125,80,130,68,134,143,111,155,103,167,100,182,101,170,107,158,109,98,207,107,199,120,193,131,193,142,188,162,185,184,187,168,206,151,218,139,222,127,223,113,219,103,206,122,203,133,201,144,197,180,189,147,204,136,208,125,209
6 | 1198_0_861.jpg,138,392,141,427,145,464,152,501,166,536,186,567,214,594,249,614,290,616,329,608,363,589,390,562,411,529,421,491,425,450,425,409,423,370,162,345,178,330,200,326,224,328,246,335,282,331,308,320,336,313,365,314,387,329,269,362,271,386,272,409,274,433,249,459,263,460,277,462,290,459,304,458,190,372,204,365,220,364,236,373,220,377,204,377,312,368,328,356,346,354,361,361,348,368,330,369,228,514,246,502,264,495,278,499,292,495,315,502,341,511,318,530,297,538,281,539,266,537,248,529,239,513,265,507,278,509,293,507,331,511,294,516,279,518,265,516
7 | 137341995_e7c48e9a75.jpg,-4,129,1,178,10,231,30,276,59,314,98,349,135,375,171,398,205,402,242,387,284,358,327,327,366,290,390,240,398,185,394,126,391,66,10,77,23,45,52,23,92,14,130,21,180,13,209,-5,248,-6,284,2,317,19,157,67,158,93,160,121,161,150,129,184,152,188,176,190,197,182,216,170,47,102,65,81,91,78,123,91,95,100,69,105,211,72,232,54,258,48,283,58,263,70,237,74,119,265,134,240,156,224,178,226,199,216,237,224,271,238,249,289,218,311,194,316,170,318,144,303,130,265,161,244,182,241,204,238,260,241,210,281,188,286,166,286
8 | 1383023626_8a49e4879a.jpg,0,48,-6,94,-10,141,-8,186,6,228,26,265,49,299,78,325,110,334,140,328,166,303,191,276,218,248,245,219,265,187,278,152,287,114,55,64,84,58,113,61,141,70,164,85,216,90,236,79,258,75,278,75,291,84,185,113,182,145,180,176,176,208,127,198,145,210,164,221,185,214,200,203,80,96,101,99,119,101,135,107,116,107,98,104,214,112,233,110,249,111,263,112,248,117,232,116,70,216,104,222,135,226,152,231,168,228,185,225,201,225,180,263,158,282,140,286,121,283,93,264,78,221,132,232,149,235,165,233,194,230,163,263,145,268,128,265
9 | 144044282_87cf3ff76e.jpg,122,104,121,116,121,129,122,142,125,155,130,167,137,178,147,187,159,190,174,189,189,183,202,175,213,165,220,152,223,138,225,123,227,108,128,96,134,92,141,92,149,95,156,98,174,96,182,94,191,93,200,94,207,99,163,105,162,113,160,121,159,130,151,135,155,137,160,139,165,138,171,137,136,103,141,101,146,102,151,105,146,105,141,105,179,106,183,104,189,104,195,106,189,108,184,108,144,152,149,149,155,148,159,150,165,149,172,151,181,154,172,158,164,159,158,160,154,159,148,157,147,152,155,153,159,154,165,154,178,154,164,153,159,154,154,153
10 | 152601997_ec6429a43c.jpg,109,66,108,98,111,132,118,169,127,203,141,236,157,264,180,281,207,288,240,288,278,279,312,261,342,236,362,200,373,161,377,118,378,75,110,40,124,28,143,26,163,29,184,34,213,28,240,19,268,17,297,23,322,36,196,68,192,93,187,118,182,143,172,164,181,168,190,170,202,167,216,166,132,70,142,61,159,62,174,77,158,80,141,79,242,75,255,57,274,56,292,64,276,75,257,78,165,223,173,209,183,201,193,203,204,199,224,205,247,220,226,227,209,231,197,231,187,230,177,228,172,222,184,217,194,217,205,216,239,220,208,216,197,217,187,216
11 | 1549040388_b99e9fa295.jpg,149,89,139,123,136,159,139,195,146,234,156,274,168,310,184,343,214,360,256,361,303,347,350,326,393,298,423,259,439,212,448,163,453,112,144,39,154,23,173,19,194,22,212,33,261,33,291,22,324,22,356,34,380,55,227,74,219,105,210,135,200,167,187,181,197,190,210,197,229,193,248,190,160,72,172,62,190,64,206,79,188,79,171,78,285,83,302,70,322,71,343,83,323,87,303,86,174,222,182,219,197,221,212,227,231,226,264,232,301,241,264,262,232,269,211,267,194,261,181,245,182,225,196,232,212,237,231,238,292,243,232,251,212,248,196,242
12 | 1878519279_f905d4f34e.jpg,144,178,145,191,149,205,157,220,166,233,175,245,184,257,194,267,205,270,221,267,238,257,257,245,271,229,281,210,284,189,283,167,281,144,142,150,145,143,152,142,159,144,167,148,184,145,196,137,210,132,224,132,239,137,177,163,177,174,176,185,175,196,172,207,177,209,183,209,190,206,198,203,150,168,154,160,162,159,170,166,163,170,155,171,202,160,207,149,216,148,226,152,219,158,209,161,177,238,176,227,180,221,186,221,191,219,204,222,218,228,210,238,199,243,193,245,187,246,182,244,180,236,183,228,188,227,193,226,213,228,196,234,190,235,185,236
13 | 2046713398_91aaa6fe1c.jpg,67,201,69,233,74,264,81,294,93,322,112,347,138,365,169,375,199,377,224,370,243,352,256,330,264,306,271,281,277,257,282,232,281,208,107,177,124,164,147,158,170,162,191,172,230,175,244,167,260,161,276,164,286,176,209,197,211,217,213,237,216,257,189,270,199,275,210,278,220,274,227,269,130,199,144,190,160,190,173,203,158,206,142,206,228,204,242,191,257,191,267,201,258,208,243,207,160,310,181,303,198,298,208,301,217,298,228,302,238,306,227,318,217,324,207,325,195,325,179,321,167,310,197,310,208,311,217,309,231,307,216,309,207,311,197,310
14 | 2173711035_dbd53b4f9f.jpg,28,127,29,172,34,214,42,255,58,292,82,326,110,355,140,378,173,382,203,373,227,350,250,324,272,297,288,264,301,229,313,192,320,153,66,145,93,132,125,134,155,142,182,154,221,155,246,146,271,141,295,140,312,152,199,179,198,211,198,245,198,277,158,275,174,283,191,291,208,285,221,277,98,173,118,170,138,170,155,179,136,182,116,182,230,180,250,173,268,173,282,176,268,185,250,184,124,302,149,302,171,302,185,306,200,302,213,303,228,300,211,319,197,332,181,336,166,335,145,324,133,305,169,312,184,315,199,312,221,304,198,311,183,315,168,313
15 | 2210514040_6b03ff2629.jpg,265,271,264,276,264,281,263,287,263,292,264,296,267,300,270,304,275,306,280,306,286,305,292,303,297,300,301,296,304,292,307,287,310,281,269,270,272,268,276,268,279,269,282,271,291,273,295,272,299,273,303,274,306,276,286,275,284,278,283,282,282,285,278,286,279,287,281,288,283,288,285,288,271,274,274,274,277,274,279,276,276,276,273,275,292,277,295,277,298,277,300,279,298,279,295,278,271,292,274,291,277,291,279,292,282,292,285,294,287,296,284,298,280,299,277,298,275,298,273,295,272,292,277,293,279,294,281,294,286,296,281,295,278,295,276,294
16 | 2322901504_08122b01ba.jpg,195,202,191,219,187,236,191,254,198,272,207,290,214,308,222,326,238,335,262,335,291,327,319,317,344,303,362,281,368,254,370,226,370,200,192,167,195,159,203,160,211,165,219,173,243,171,261,163,280,160,300,166,316,177,229,189,224,203,217,216,211,230,206,241,213,246,222,250,232,247,244,245,201,189,206,182,216,183,225,193,215,194,205,194,259,196,270,186,283,187,296,196,283,200,271,199,207,266,208,262,216,260,224,263,233,262,254,266,281,271,257,287,237,292,228,291,219,289,211,280,212,267,218,268,225,269,234,269,273,272,235,279,226,277,219,274
17 | 2327253037_66a61ea6fe.jpg,118,198,123,225,130,253,140,279,156,304,178,324,204,338,231,345,257,341,276,331,292,311,303,290,313,267,320,244,321,222,317,202,313,182,150,177,162,162,182,152,203,149,222,155,248,147,261,139,276,138,289,145,295,157,238,168,243,186,247,203,252,221,219,234,232,238,246,242,259,236,269,230,174,185,185,180,196,178,207,180,199,182,188,184,257,175,266,171,275,171,285,175,276,176,267,176,190,260,211,256,233,254,247,256,259,252,274,253,288,255,275,273,262,284,248,286,235,285,212,278,196,261,233,260,247,260,259,258,282,257,261,274,247,278,234,276
18 | 2328398005_d328a70b4c.jpg,13,183,8,224,6,267,11,308,31,343,61,372,96,393,134,408,169,414,196,408,213,387,228,365,242,340,255,314,264,287,269,262,270,236,98,160,124,152,152,152,179,158,202,172,234,188,248,183,264,182,276,185,282,196,214,206,216,228,218,251,221,275,175,286,189,293,203,299,215,298,224,291,121,192,139,188,154,190,166,204,150,204,135,201,225,216,241,209,254,211,261,223,252,226,238,222,131,327,162,326,185,326,194,331,202,329,209,333,213,338,204,349,196,352,187,352,177,351,157,345,139,330,182,335,192,337,199,337,208,338,199,339,191,338,181,337
19 | 2370961440_6bc8ce346c.jpg,66,132,68,160,72,187,82,214,99,234,124,247,152,254,178,259,203,260,222,255,234,241,245,225,251,206,256,186,256,166,252,146,245,127,108,95,123,84,142,78,162,78,181,85,203,89,215,87,228,89,237,93,244,102,196,109,199,119,204,129,208,140,182,161,192,164,202,165,209,164,216,161,130,113,142,106,153,106,164,114,153,115,141,115,207,118,217,112,227,114,233,122,226,122,216,120,167,199,186,190,199,184,205,186,211,184,219,189,225,198,219,204,212,205,206,206,199,206,186,205,173,199,199,195,205,194,210,194,221,197,211,195,206,195,200,195
20 | 2382SJ8.jpg,50,100,49,112,50,124,52,137,56,149,63,159,71,169,82,175,95,178,109,177,121,171,131,164,138,154,144,143,147,130,149,118,150,105,58,94,64,86,74,83,84,84,94,86,108,86,118,84,128,84,136,88,142,96,100,96,99,103,98,109,98,116,83,121,90,123,97,125,105,124,112,124,66,99,73,96,79,96,85,99,79,99,73,99,115,101,121,97,127,97,132,101,127,101,121,101,73,141,80,133,90,130,97,132,105,131,114,136,122,145,113,152,104,155,96,155,88,153,79,149,76,140,89,136,97,136,104,137,118,144,104,147,96,147,89,146
21 | 252418361_440b75751b.jpg,265,146,263,166,266,188,273,211,282,233,293,252,305,272,318,287,340,291,367,288,399,277,430,261,455,237,472,207,476,172,472,134,468,99,257,114,259,105,266,100,277,99,288,102,308,94,329,84,352,78,376,81,398,91,300,117,297,135,293,153,289,171,289,189,297,193,307,194,317,188,329,184,272,129,277,120,287,117,297,124,288,129,278,131,341,114,348,103,361,100,375,105,364,112,352,114,300,232,298,226,305,222,314,223,324,219,344,219,368,220,350,231,333,240,322,242,313,243,304,239,302,231,308,229,316,229,327,227,361,221,328,227,318,229,309,228
22 | 262007783_943bbcf613.jpg,123,137,128,176,137,215,150,250,171,278,200,302,230,318,262,329,289,326,307,315,320,291,330,268,341,244,352,220,360,195,363,169,361,145,177,107,197,89,224,82,252,84,278,95,326,99,338,91,351,88,363,89,367,99,305,131,310,159,317,187,324,215,283,225,297,229,310,233,321,229,328,221,212,134,227,121,246,121,260,135,246,140,227,140,315,134,328,119,344,119,351,130,344,137,329,137,247,258,271,253,292,250,302,252,311,248,317,248,321,250,317,260,312,269,302,273,289,273,270,269,255,258,291,259,301,260,311,256,316,252,312,254,302,258,292,258
23 | 2633371780_45b740b670.jpg,101,174,117,201,137,226,159,248,186,268,213,285,236,301,261,312,287,307,313,288,335,260,353,228,365,192,368,153,358,113,345,73,328,35,83,144,86,130,100,122,117,118,138,117,159,99,178,80,200,66,226,57,255,58,163,134,172,163,180,192,189,221,180,228,194,232,210,232,223,220,235,208,112,157,118,147,131,141,147,143,136,151,124,157,205,116,212,103,226,95,245,92,234,103,220,111,204,260,208,257,216,252,227,251,236,242,257,233,281,222,272,247,258,263,248,270,236,273,222,270,209,260,221,257,232,255,242,249,276,227,248,250,238,256,227,257
24 | 2647088981_60e9fe40cd.jpg,93,213,99,226,106,238,113,250,124,260,138,266,153,268,169,269,183,266,192,261,197,252,201,242,201,229,199,216,196,204,192,192,188,182,107,201,112,192,122,186,133,184,144,185,155,181,161,175,168,171,177,171,182,176,153,190,158,197,162,204,167,211,156,223,162,222,168,222,171,219,174,216,122,202,126,197,132,195,138,197,134,200,128,202,163,189,166,183,172,181,176,183,174,186,169,188,151,247,158,239,165,234,170,233,173,231,179,230,184,233,182,235,178,237,175,239,170,241,162,244,153,245,166,237,171,236,175,234,182,233,176,234,172,236,168,238
25 | 2711409561_a0786a3d3d.jpg,179,159,180,186,182,211,185,238,191,264,202,285,217,306,233,325,256,332,283,331,309,318,331,302,351,281,367,255,377,225,381,194,383,161,180,143,188,131,206,132,222,139,238,150,274,152,294,145,314,141,336,144,352,154,258,174,258,196,258,217,257,238,237,241,247,248,258,253,270,250,281,245,198,164,208,157,224,160,235,173,221,176,206,174,291,177,303,167,317,167,330,174,318,180,304,180,222,276,234,273,246,273,257,276,270,273,283,273,298,278,283,291,268,298,256,299,243,297,231,290,227,277,246,280,257,281,269,280,291,279,269,285,257,287,245,285
26 | 2722779845_7fcb64a096.jpg,71,128,63,151,58,173,56,196,57,218,63,240,74,257,90,270,111,280,135,287,161,288,187,286,208,277,224,262,238,244,249,224,259,203,89,107,103,95,123,93,142,99,158,111,181,119,205,120,228,127,245,141,251,164,162,140,156,149,149,158,142,168,122,184,129,189,136,194,146,197,156,200,101,129,114,123,129,126,138,141,124,141,110,137,185,159,202,152,216,158,224,173,212,173,197,167,95,215,109,210,121,210,128,216,137,217,149,228,163,244,143,244,129,240,120,236,111,232,103,225,101,216,118,219,126,223,135,226,156,240,133,228,124,225,116,221
27 | 2795838930_0cc5aa5f41.jpg,234,92,228,107,222,122,218,139,218,154,222,169,229,182,238,193,253,201,269,206,287,204,302,199,315,191,325,181,334,169,341,156,347,143,250,87,260,82,271,83,282,88,290,98,302,104,315,105,328,108,339,117,342,129,292,112,287,123,283,132,278,143,261,144,267,149,273,154,281,155,289,156,256,101,263,101,270,104,274,110,267,108,260,105,307,124,315,123,322,127,327,132,320,131,313,128,243,155,254,157,263,159,269,163,276,165,284,171,293,176,280,177,269,176,262,173,256,169,249,163,246,157,261,164,267,168,274,170,290,176,272,170,265,168,259,164
28 | 2902323565_100017b63c.jpg,79,219,77,246,78,273,82,300,88,325,99,350,114,372,132,388,155,394,177,391,195,377,211,359,225,339,236,317,244,294,249,270,252,245,97,207,109,195,126,192,143,195,159,204,187,208,203,204,220,205,235,212,242,228,170,229,168,249,166,267,165,287,143,296,152,300,162,305,172,303,181,300,112,225,123,218,135,221,145,233,133,234,120,232,189,239,201,230,213,231,222,241,213,244,200,243,123,322,136,320,151,320,161,323,171,321,184,322,195,324,184,341,171,350,160,351,149,349,135,339,128,324,150,328,160,330,170,328,190,325,171,334,161,336,151,334
29 | 2902760364_89c50bde40.jpg,69,235,72,263,78,290,84,316,95,340,112,360,132,378,151,394,172,398,193,393,214,377,235,359,253,338,264,313,269,284,272,255,274,224,80,235,95,225,114,225,133,230,152,237,188,235,206,226,224,220,243,218,259,226,169,256,169,274,168,293,168,312,150,318,159,321,169,324,178,321,187,317,106,250,118,242,134,243,145,257,132,260,116,259,195,255,206,240,222,238,235,245,225,255,209,257,132,340,147,341,160,340,168,342,176,339,190,341,206,339,190,353,177,360,168,361,159,360,147,354,139,343,160,347,168,348,177,347,199,342,177,347,168,349,160,347
30 | 2956581526_cd803f2daa.jpg,79,65,77,81,78,96,80,111,83,126,88,139,97,150,109,159,122,162,138,162,156,158,172,150,183,139,190,124,194,108,198,92,201,74,83,52,89,45,99,43,109,45,119,51,144,52,155,49,168,48,179,52,187,60,130,66,129,77,127,89,125,100,116,105,120,108,126,110,132,109,139,107,94,63,101,59,109,61,116,68,108,68,99,66,149,70,157,64,165,64,172,70,165,71,156,71,105,120,114,120,120,119,126,121,132,121,140,123,151,126,139,130,130,131,124,130,118,129,112,127,109,122,120,124,126,125,131,126,147,126,131,125,125,124,119,123
31 | 297448785_b2dda4b2c0.jpg,25,194,27,239,34,283,44,326,61,368,83,407,109,441,143,466,176,477,210,473,243,453,271,422,295,387,316,347,332,304,345,259,350,215,53,139,71,113,101,102,134,103,165,112,228,114,259,107,290,109,317,122,332,148,194,160,192,181,190,202,187,223,161,256,173,262,185,265,200,265,213,263,89,170,107,156,133,156,153,174,132,179,106,179,234,178,254,162,280,165,298,183,280,190,255,187,141,353,147,316,169,300,180,305,195,305,215,329,222,372,211,391,192,403,176,403,161,398,144,379,151,351,168,323,179,325,193,328,211,367,192,378,178,377,165,373
32 | 299733036_fff5ea6f8e.jpg,122,149,121,163,119,177,119,192,124,205,134,216,146,226,158,233,172,237,182,236,189,229,194,219,198,210,202,200,206,191,209,182,211,174,149,144,159,140,169,140,179,143,187,149,199,153,205,152,211,152,216,155,217,161,191,161,192,169,193,176,194,183,178,188,182,191,187,193,190,193,193,192,157,155,163,153,169,155,173,160,168,160,162,158,194,167,200,164,205,166,207,170,204,171,198,170,163,204,172,199,180,199,184,201,188,201,190,206,190,213,187,214,184,214,180,213,175,211,169,208,166,204,179,204,183,206,186,207,188,211,186,207,182,206,178,204
33 | 303808204_1f744bc407.jpg,167,137,170,168,176,199,182,230,192,260,210,284,233,304,260,320,291,323,321,315,344,295,363,272,377,245,383,215,386,185,389,155,390,126,187,130,204,118,226,115,249,116,270,123,292,121,312,115,334,112,356,114,373,126,282,148,283,167,284,186,286,205,263,221,274,224,287,226,298,222,308,218,213,147,226,141,241,142,253,153,240,155,224,155,311,151,323,140,338,138,351,144,340,152,325,153,248,259,263,249,276,244,287,247,298,243,312,246,326,253,314,262,301,268,289,270,278,269,264,267,255,258,277,255,288,256,299,253,319,252,300,254,288,257,277,256
34 | 3074791551_baee7fa0c1.jpg,187,160,186,179,188,198,193,216,200,232,210,248,221,261,233,271,247,275,263,275,278,266,291,254,303,239,312,222,320,205,326,187,329,168,196,147,208,142,221,143,234,148,246,156,277,158,289,152,302,149,315,151,324,158,258,169,256,185,254,200,253,216,238,219,245,222,252,226,260,223,268,221,209,163,218,159,229,160,238,169,227,170,216,169,280,172,290,165,301,165,309,170,301,175,290,175,224,237,234,236,244,235,251,238,258,236,267,238,278,241,267,250,257,253,250,253,242,252,233,247,229,238,243,240,250,242,258,241,273,242,258,241,250,242,243,240
35 | 3152653555_68322314f3.jpg,88,63,84,83,82,103,82,124,85,144,94,162,106,178,118,192,133,197,146,197,159,187,171,175,182,163,193,150,202,136,210,122,216,106,112,62,126,58,141,58,155,63,167,72,186,78,197,76,207,76,216,79,219,88,173,91,171,104,169,117,166,130,148,131,154,136,159,140,165,140,169,139,126,81,136,79,145,81,150,89,143,89,133,87,182,98,192,95,201,97,205,102,199,105,191,103,118,143,133,145,147,146,152,149,159,150,166,154,171,158,160,167,151,169,143,168,137,166,127,157,123,145,144,150,150,153,156,154,166,158,154,161,147,160,141,157
36 | 3264867945_fe18d442c1.jpg,194,167,194,185,195,204,199,222,206,239,216,255,228,269,243,278,259,279,275,276,289,267,301,254,311,238,317,222,320,204,320,186,319,169,201,148,207,135,219,129,234,129,247,134,266,134,278,129,292,129,304,135,310,147,257,152,257,161,257,170,257,179,246,197,252,198,257,199,262,198,268,197,216,157,222,151,232,150,241,158,232,160,222,161,272,158,279,151,289,151,296,158,289,161,280,160,236,235,242,225,251,219,257,222,261,220,271,225,277,235,271,246,263,251,257,251,251,251,242,246,241,234,251,228,257,228,261,228,273,234,262,239,257,240,251,239
37 | 3273658251_b95f65c244.jpg,220,85,221,110,224,135,229,159,237,182,251,202,270,219,290,234,315,238,339,235,356,222,371,204,384,183,391,161,396,137,399,113,399,89,238,72,250,59,269,54,287,55,305,60,325,60,342,54,360,53,377,59,388,72,315,81,316,96,316,112,317,128,296,140,306,142,316,145,327,142,336,140,258,86,268,80,279,80,289,87,278,89,267,89,339,87,349,80,360,80,370,86,361,90,350,89,281,176,293,167,306,162,316,165,326,162,338,168,349,177,338,187,327,192,316,192,306,191,294,186,288,177,306,175,316,176,326,175,343,177,326,174,316,175,306,174
38 | 3298715079_5af7c78fcb.jpg,194,112,187,128,182,146,181,165,182,183,187,203,193,220,202,235,218,243,240,244,269,241,297,233,321,220,337,201,345,176,349,150,352,125,193,86,197,80,204,79,211,80,218,84,236,84,251,80,269,80,286,86,300,96,222,102,216,113,209,124,202,135,198,152,204,155,210,158,219,157,229,156,197,109,201,101,209,101,217,107,209,109,201,110,250,109,258,102,268,103,280,109,268,111,258,111,197,185,199,178,204,173,210,175,216,174,230,181,249,190,230,196,217,198,210,198,204,196,199,192,200,185,205,184,210,184,217,184,242,189,217,186,210,185,205,185
39 | 3325611505_ddc7beffa1.jpg,134,118,135,133,137,149,139,164,144,178,152,191,161,202,172,211,185,214,197,212,207,204,215,192,224,179,229,165,233,150,235,135,236,120,142,119,149,112,159,110,168,113,178,117,195,118,204,113,214,110,224,112,229,120,186,131,186,142,186,153,187,164,176,169,181,170,186,172,191,171,196,169,153,129,159,125,167,126,173,132,166,135,158,134,198,133,204,127,212,125,218,130,213,134,205,135,168,186,175,184,181,182,186,184,190,182,196,184,202,186,196,192,191,194,186,195,180,194,174,191,171,186,180,187,185,188,190,187,199,187,191,187,186,188,181,187
40 | 3362762930_24f76cb89c.jpg,106,142,114,175,123,209,135,240,153,268,176,291,203,310,232,319,260,318,285,307,307,290,325,270,340,245,350,217,353,186,353,155,351,124,131,123,148,110,170,107,195,109,219,116,268,116,286,105,305,95,325,93,342,102,243,134,247,155,250,176,254,198,224,209,239,210,255,213,268,206,281,199,160,146,174,136,191,134,207,143,192,148,175,149,274,135,285,121,301,118,317,124,305,131,289,134,192,239,216,233,240,227,256,227,269,221,290,220,311,221,294,248,275,262,259,268,243,268,218,261,200,240,241,237,257,237,270,232,304,225,272,248,258,252,242,253
41 | 343583208_e986824d77.jpg,-8,339,2,368,14,398,31,424,55,442,86,452,118,457,151,458,179,451,199,438,209,419,217,398,220,374,219,351,217,327,209,302,197,280,21,294,34,275,55,263,79,260,103,263,138,257,150,246,164,241,178,240,188,249,128,283,136,297,144,309,153,322,128,349,141,348,154,346,163,342,171,336,54,304,66,291,81,287,98,296,85,303,68,307,147,284,156,269,169,266,180,273,173,281,160,283,120,402,136,383,152,370,164,370,173,364,186,366,195,377,193,390,185,399,175,403,163,407,144,407,129,399,154,385,165,382,175,378,189,377,180,381,170,385,159,387
42 | 3461016494_56cce9c984.jpg,100,187,102,204,106,221,112,236,119,250,129,263,140,272,154,279,169,280,184,276,196,266,207,253,215,238,221,222,223,205,224,187,224,169,108,178,116,170,127,169,138,170,149,172,171,167,182,161,193,156,205,156,214,161,163,183,164,195,165,207,166,220,153,225,160,227,166,229,174,226,180,222,122,192,130,189,138,188,146,189,139,192,131,193,181,184,189,181,197,179,204,179,198,184,190,185,144,245,154,242,161,239,166,241,172,238,179,239,188,240,180,247,174,252,167,253,161,253,154,251,148,245,161,245,167,245,173,244,184,241,173,242,166,244,161,244
43 | 348272697_832ce65324.jpg,127,222,127,246,128,269,130,292,136,315,151,333,172,345,198,352,226,356,252,354,274,346,290,332,299,314,304,294,306,274,309,254,310,235,156,197,169,182,189,176,210,176,229,182,251,184,268,183,284,186,298,195,304,210,241,195,241,206,242,217,243,229,219,250,228,252,237,255,246,254,254,253,176,202,187,195,199,195,209,203,198,203,187,203,258,207,269,202,280,204,288,213,279,212,268,210,191,287,209,279,224,276,233,279,242,278,253,285,263,296,250,297,239,296,230,295,220,294,206,291,197,287,223,284,232,286,241,286,258,294,240,287,231,287,222,285
44 | 3534188114_2108895291.jpg,189,153,188,170,189,187,192,203,199,218,210,231,224,241,237,250,251,253,264,251,274,241,282,230,289,219,295,207,300,194,303,180,303,167,213,143,222,136,234,135,246,139,256,145,274,146,282,143,290,142,297,144,300,152,265,156,265,165,266,174,267,184,249,188,256,191,262,194,267,193,271,192,224,154,230,151,237,152,243,157,236,158,229,157,273,161,280,157,287,158,291,162,286,164,280,163,225,204,238,201,249,201,257,204,264,203,271,205,276,210,269,219,261,222,254,222,245,220,235,214,229,204,248,205,256,207,263,207,273,210,262,216,255,217,247,215
45 | 3534189272_8ef88ba368.jpg,212,152,210,165,210,178,211,192,214,205,218,218,226,229,235,237,248,241,262,241,275,237,287,230,296,220,304,208,309,195,313,182,315,168,219,138,225,132,233,129,242,131,249,137,270,136,279,133,289,134,298,139,303,147,257,147,255,156,254,166,252,175,242,180,246,184,252,186,258,184,265,182,227,149,232,144,239,145,246,150,239,150,232,150,273,153,280,149,287,150,292,155,286,155,279,154,230,194,239,194,247,194,252,195,258,194,268,196,278,199,268,207,258,210,251,210,245,209,238,205,234,196,246,197,252,198,258,198,275,199,258,205,252,205,246,204
46 | 3555944509_7b477069c6.jpg,27,131,27,166,31,200,36,233,48,263,66,290,88,313,116,331,147,337,181,336,211,321,233,296,254,267,269,234,278,199,287,162,292,122,43,98,60,86,84,84,107,89,128,98,180,97,206,86,233,81,260,85,279,101,153,129,152,157,151,184,150,211,124,219,136,226,150,232,165,228,179,223,70,127,86,119,104,120,119,133,101,135,84,135,192,134,210,120,229,119,245,129,231,137,211,137,105,259,123,254,140,253,151,259,164,256,184,261,205,268,184,283,163,288,149,287,136,283,121,272,112,260,139,265,151,269,164,269,197,268,163,273,150,272,138,267
47 | 3574737496_6ee8207045.jpg,186,149,184,166,183,184,185,201,192,217,204,230,217,240,232,248,247,250,261,248,271,240,279,229,287,218,293,206,296,192,297,178,297,165,206,138,217,130,230,130,242,133,254,139,269,140,278,138,287,137,296,141,298,150,262,152,262,161,264,169,265,178,245,184,252,186,259,189,266,188,271,187,222,150,229,147,236,148,241,153,235,153,228,152,269,157,276,154,282,155,286,159,282,160,275,159,226,201,238,199,250,198,256,201,262,200,269,203,273,207,266,219,258,223,252,222,245,220,235,213,229,202,249,202,255,204,261,204,270,208,260,213,254,214,247,212
48 | 362167809_d5a5dcbfdb.jpg,111,150,116,170,121,190,129,209,142,225,160,237,183,244,207,245,228,242,244,235,254,221,260,205,263,187,263,170,261,153,258,137,254,121,134,125,144,114,158,106,174,105,189,107,210,103,220,97,232,94,242,96,248,105,204,121,208,132,213,142,217,153,199,170,207,171,216,172,221,168,226,164,153,136,160,128,170,126,180,131,171,135,161,137,216,124,222,116,231,114,238,119,233,124,224,125,186,202,200,196,211,190,218,191,222,187,229,188,237,189,232,196,227,200,221,202,215,204,202,204,191,201,212,197,219,196,223,194,233,190,225,193,220,196,213,197
49 | 363149951_8be04dc6c0.jpg,88,160,90,181,92,201,94,222,99,241,110,257,124,271,141,282,160,285,178,282,194,272,208,259,219,243,224,224,226,204,230,185,233,165,99,152,108,143,121,139,135,140,148,145,174,147,187,143,201,143,214,147,222,157,160,159,160,170,160,181,160,192,143,203,151,205,159,208,168,206,175,204,114,160,122,155,132,155,141,162,132,163,122,163,179,163,188,157,198,157,206,163,198,165,188,164,125,226,137,222,149,221,158,223,168,222,181,224,192,230,181,243,168,247,158,248,148,246,136,240,130,226,149,226,158,227,168,227,188,230,168,239,159,240,149,238
50 | 3638950581_3387685d3a.jpg,164,135,160,150,158,165,162,180,168,195,173,209,176,225,179,242,190,249,208,251,228,246,248,236,268,222,282,205,292,183,297,159,300,135,162,112,164,105,169,104,176,107,181,112,203,111,217,106,233,104,249,108,261,118,190,127,185,141,179,154,174,167,171,173,176,178,183,181,192,178,202,176,168,127,172,122,181,122,187,129,179,130,172,129,217,131,225,124,234,124,243,130,234,132,225,132,175,189,175,188,181,189,187,192,195,191,212,193,233,195,214,210,197,216,188,216,181,213,176,203,177,191,181,194,187,196,195,196,229,196,196,207,188,207,182,203
51 | 3646828311_bfeb429ef7.jpg,103,188,107,205,113,223,120,240,131,253,147,265,163,275,179,283,195,283,210,276,221,262,232,249,242,233,248,216,249,198,248,179,246,159,108,174,115,166,126,162,139,162,151,164,181,160,191,152,203,146,216,144,230,149,170,178,172,193,175,208,178,224,163,230,172,231,181,232,189,228,196,224,125,185,132,180,141,179,151,183,143,185,133,187,190,176,198,168,207,165,217,167,210,172,200,174,152,241,163,242,173,241,182,242,190,239,202,236,214,232,206,246,195,255,186,257,177,257,165,253,156,243,174,244,183,244,192,242,210,235,193,246,185,249,175,249
52 | 3689162471_5f9ffb5aa0.jpg,237,83,235,94,233,106,233,118,235,129,240,140,246,150,254,158,263,162,272,162,281,156,289,149,296,142,301,133,305,124,309,115,312,105,250,85,257,84,264,86,270,89,276,93,291,96,298,95,304,95,309,96,312,101,282,101,280,109,278,117,277,124,267,125,270,127,274,130,278,129,281,128,256,94,261,92,266,94,269,98,264,98,259,97,289,103,294,101,299,102,302,105,298,106,293,105,258,133,263,136,269,137,272,138,276,138,280,139,283,140,278,144,274,145,270,145,266,144,261,140,260,135,268,139,271,140,275,140,281,141,275,140,271,140,267,139
53 | 3718903026_c1bf5dfcf8.jpg,125,220,124,245,123,269,125,295,134,318,151,337,169,354,189,368,213,373,238,369,259,355,277,340,294,323,305,302,310,278,312,255,315,231,146,206,156,195,171,190,187,193,201,201,247,203,262,196,278,195,293,201,301,214,223,223,222,241,221,259,220,277,195,275,206,281,219,288,232,283,242,278,164,225,176,225,185,225,195,226,185,226,175,226,249,228,260,229,270,229,280,230,270,230,260,229,163,293,182,291,201,294,216,298,233,296,251,295,268,298,251,323,231,332,214,333,198,331,179,320,168,295,201,298,216,301,232,300,263,300,232,323,215,324,199,320
54 | 3790616528_297c0ac935.jpg,51,192,52,216,54,240,59,263,70,285,87,301,108,314,130,323,150,325,169,321,185,309,201,294,211,276,217,255,220,233,220,212,221,193,77,182,91,173,108,171,125,175,142,180,164,182,179,178,192,174,207,176,215,186,154,191,156,203,158,216,160,228,135,240,145,242,154,244,163,242,170,240,96,191,106,189,116,188,125,194,116,195,106,194,172,195,182,190,192,190,200,194,192,197,182,197,113,270,128,264,141,261,150,263,160,261,170,266,181,272,170,277,159,279,150,279,140,277,129,275,119,269,141,268,150,269,160,269,176,272,159,270,150,270,141,269
55 | 3855944735_e252959937.jpg,21,260,25,288,29,315,39,339,55,358,75,371,102,379,131,377,161,374,187,370,205,362,218,346,224,326,229,304,229,282,225,262,220,244,63,205,77,186,97,173,120,170,139,178,169,179,184,172,200,173,214,181,222,196,155,193,159,204,163,215,166,226,136,253,148,254,161,256,172,253,182,250,85,217,98,208,110,204,123,209,111,212,99,214,176,205,190,202,200,202,207,209,198,207,189,206,111,306,128,284,149,277,163,278,175,275,189,280,198,298,189,309,176,314,164,316,149,316,130,315,118,303,149,284,164,283,175,282,193,297,175,305,164,306,149,307
56 | 3856149136_d4595ffdd4.jpg,74,161,79,191,88,219,95,247,102,275,117,300,137,321,162,338,189,343,216,338,238,321,255,299,268,273,274,244,280,215,285,185,287,156,92,155,106,141,125,137,146,138,167,141,193,141,215,138,236,136,256,140,270,152,182,159,183,178,185,197,186,217,162,231,174,234,186,237,197,234,208,230,114,162,125,155,138,155,151,164,138,164,125,165,212,163,224,154,237,155,250,161,238,164,225,163,144,270,160,263,175,258,185,261,195,258,211,263,227,268,212,281,197,287,186,289,175,287,160,281,151,271,175,269,185,270,196,269,221,269,196,271,185,273,175,271
57 | 3872768751_e60d7fdbd5.jpg,268,110,271,124,277,136,283,148,290,160,301,170,314,176,328,180,341,179,353,174,362,165,371,155,377,142,378,128,378,114,377,101,376,88,279,96,285,89,294,84,305,82,314,84,330,81,337,75,346,72,355,72,364,75,324,94,326,102,328,110,330,118,321,129,327,130,332,130,337,127,341,123,292,104,297,100,304,98,311,101,305,104,298,105,338,95,343,90,349,89,355,91,350,94,344,95,311,147,321,144,329,142,334,142,339,140,347,139,356,137,349,146,343,151,337,153,332,154,323,152,314,147,329,145,335,145,340,143,353,139,341,146,336,148,330,149
58 | 529447797_0f9d2fb756.jpg,78,173,81,219,89,266,105,312,128,353,153,392,180,427,210,451,247,457,282,452,311,431,340,398,366,360,386,318,401,272,408,223,410,177,102,163,123,138,155,133,188,141,218,156,260,153,293,139,327,132,361,140,384,164,240,188,241,225,242,263,242,300,209,307,224,314,241,321,258,315,274,308,138,185,157,172,183,174,202,194,180,199,155,198,281,193,302,172,327,171,346,184,330,197,305,198,185,353,205,348,226,345,240,350,255,346,279,351,299,357,280,389,258,408,242,410,226,406,207,387,193,357,226,356,240,359,256,357,291,361,257,383,241,387,226,382
59 | 57635685_d41c98f8ca.jpg,86,118,87,153,87,189,93,225,113,253,141,273,173,287,206,297,235,297,255,288,262,267,268,246,273,224,278,200,280,179,280,156,274,138,155,113,174,101,198,96,221,100,238,112,248,114,258,107,269,106,279,109,281,120,246,131,251,145,257,159,263,174,224,190,237,194,250,199,261,195,269,189,175,126,188,122,199,123,209,131,199,131,187,130,248,134,258,128,267,129,272,135,266,136,257,135,197,230,219,221,237,217,247,221,254,219,260,222,260,231,258,241,252,245,244,245,234,243,217,238,204,228,236,226,247,227,253,226,256,229,252,232,246,232,235,230
60 | 809285949_6889026b53.jpg,27,108,30,123,34,138,40,152,48,165,58,176,69,185,81,195,97,196,115,192,133,183,151,172,165,157,173,138,175,118,174,95,170,74,26,95,30,88,39,86,50,86,60,90,75,87,88,77,103,71,121,68,137,73,71,99,71,111,72,123,72,136,65,142,72,145,80,146,89,142,98,137,39,107,44,103,52,101,62,104,54,107,46,109,97,97,103,91,112,89,122,90,114,95,105,96,67,160,70,156,77,154,85,155,92,151,105,150,119,150,108,160,97,167,89,169,81,169,73,166,71,160,78,159,86,159,94,157,115,151,94,157,87,159,79,159
61 | 92053278_be61a225d2.jpg,145,83,143,112,143,140,148,167,158,193,174,215,193,235,212,257,235,265,261,264,290,255,319,243,343,222,358,195,364,163,368,129,370,96,157,59,167,47,183,43,200,45,217,52,264,54,283,47,303,47,322,54,338,66,236,79,234,98,230,117,227,136,213,148,222,152,231,156,243,153,254,151,174,83,186,72,200,73,214,87,199,89,184,89,274,90,286,77,302,77,315,89,302,94,287,93,198,177,212,172,224,172,235,174,247,173,266,176,289,182,266,197,247,202,234,202,223,199,210,192,204,178,224,176,235,179,247,179,282,182,247,192,235,192,224,189
62 | 96063776_bdb3617b64.jpg,162,177,165,198,170,219,175,240,182,260,195,278,212,291,233,300,257,303,278,296,296,283,310,265,317,245,321,223,323,201,325,179,324,156,178,160,188,147,204,140,222,140,238,145,264,140,277,132,292,129,307,132,316,144,252,160,253,172,254,184,255,197,238,213,247,214,255,215,264,213,271,210,198,166,209,158,220,157,230,166,220,169,209,169,272,161,281,151,292,149,301,155,294,160,283,162,222,244,232,235,246,232,256,233,265,230,277,232,287,240,278,256,266,263,256,265,245,265,232,258,227,243,246,236,256,237,265,235,282,240,265,255,255,257,245,256
63 | 97308305_4b737d0873.jpg,137,157,139,190,142,223,149,255,164,283,182,305,205,322,233,332,264,333,289,326,304,310,315,289,324,266,334,242,342,215,348,188,350,160,173,135,189,119,210,113,233,115,253,125,287,123,303,115,321,113,338,118,347,132,272,140,274,161,275,183,277,205,244,213,258,217,272,221,284,218,295,214,196,145,208,139,221,139,233,147,221,148,208,148,297,147,310,140,322,140,331,146,322,149,310,148,220,255,241,248,259,246,270,248,281,246,294,249,305,256,292,271,279,276,267,277,255,275,238,270,226,255,258,253,270,255,281,254,298,257,279,261,268,262,256,261
64 | britney-bald.jpg,52,134,54,149,56,164,60,179,65,193,73,207,81,219,92,228,106,232,121,230,136,222,150,212,160,198,167,183,171,166,174,149,174,131,54,127,61,122,71,122,81,124,90,128,115,129,125,125,137,122,148,122,159,127,102,138,102,149,102,160,101,171,92,176,97,179,103,180,109,178,115,176,66,137,73,133,81,133,89,139,80,141,72,141,122,140,130,133,139,133,146,138,139,142,130,142,83,191,91,190,99,188,104,189,110,188,120,190,130,192,120,203,110,207,104,208,98,207,91,202,87,193,98,192,104,193,110,193,126,193,110,199,104,200,98,198
65 | deeny.peggy.jpg,39,95,40,106,41,116,44,126,49,136,54,145,60,152,66,159,75,161,84,158,95,153,106,146,115,137,121,126,124,113,125,98,126,84,38,91,40,86,46,84,52,84,59,85,68,83,76,80,85,78,94,79,103,83,63,94,63,103,62,111,62,119,58,124,61,125,66,126,70,124,75,122,44,95,48,92,54,92,58,96,54,99,48,99,78,94,82,88,89,87,94,90,90,94,83,95,60,137,61,136,65,134,68,135,71,133,78,133,88,133,80,139,73,141,69,142,66,143,63,141,62,137,65,137,68,137,72,136,85,134,72,137,69,138,66,137
66 | matt-mathes.jpg,85,152,83,173,83,195,85,217,91,238,103,257,117,274,134,287,154,290,176,287,197,276,218,263,235,247,246,227,251,205,253,181,254,157,91,136,98,127,112,125,126,129,140,135,162,134,177,128,193,126,209,126,222,134,148,149,146,162,144,176,143,190,129,199,137,203,147,206,157,202,167,199,103,148,111,142,122,143,131,151,121,151,111,151,177,150,186,143,196,143,205,148,197,151,186,151,116,224,127,221,140,221,149,223,160,220,177,220,195,224,179,238,163,244,152,245,141,244,129,239,120,225,140,225,150,226,160,224,190,225,162,236,151,237,141,235
67 | person-7.jpg,32,65,33,76,34,86,34,97,37,107,41,116,50,122,61,126,72,127,83,126,95,123,107,119,115,111,118,101,120,91,122,80,122,68,39,52,45,46,53,44,61,46,68,49,82,49,90,45,98,44,106,46,112,52,74,57,74,63,74,69,74,75,67,83,70,84,74,85,78,84,82,83,47,61,51,57,58,57,63,61,57,63,51,63,87,62,93,58,98,58,103,61,99,63,93,63,55,98,63,96,70,94,75,95,80,94,86,95,94,99,86,103,79,105,74,105,69,105,62,103,58,99,70,98,74,98,79,98,91,99,79,99,74,99,69,99
68 | person.jpg,78,83,79,93,80,103,81,112,85,121,91,129,98,136,106,142,115,144,125,142,134,137,142,130,149,122,153,113,154,102,155,91,156,79,84,75,87,71,93,70,99,69,104,72,121,71,127,68,134,67,140,68,146,72,113,81,113,88,112,95,112,102,105,105,109,107,113,108,117,106,121,104,91,83,95,81,100,81,105,84,100,84,95,85,123,83,127,80,132,79,137,81,133,83,128,83,96,114,102,112,109,112,114,114,119,112,126,112,134,113,127,122,120,126,114,126,109,126,102,122,99,115,109,115,114,116,119,115,132,114,120,121,114,122,109,121
69 | person_TjahjonoDGondhowiardjo.jpg,41,92,41,103,43,114,44,125,48,134,54,142,61,148,70,152,80,154,90,153,99,149,107,143,113,135,117,126,119,115,121,104,122,93,49,83,54,79,60,78,67,79,73,82,87,81,94,79,101,78,108,80,112,84,80,90,80,96,80,102,80,108,72,115,76,116,80,116,84,116,88,115,57,90,61,87,67,87,71,91,66,92,61,92,91,91,95,88,101,88,105,91,101,93,95,93,66,130,72,127,77,125,81,126,85,125,90,127,96,130,90,132,85,134,81,134,77,133,71,133,68,130,77,130,81,130,85,130,94,130,85,128,81,128,77,128
70 | personalpic.jpg,40,109,40,122,41,135,42,148,44,161,49,174,56,186,64,196,76,199,89,198,103,192,116,183,127,173,135,160,140,145,142,129,144,112,39,103,44,97,53,96,61,98,70,100,84,100,96,97,108,96,118,100,126,107,76,109,75,121,74,132,72,143,63,144,68,148,73,151,80,149,87,146,49,109,54,106,61,106,66,110,60,111,54,111,94,112,100,108,107,109,113,112,107,114,100,114,58,162,63,160,68,160,74,163,80,161,88,163,97,165,87,169,79,171,73,171,67,170,62,167,60,162,68,164,74,166,80,165,94,165,80,165,74,165,68,164
71 |
--------------------------------------------------------------------------------
/PracticePytorch/faces/matt-mathes.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/matt-mathes.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/person-7.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/person-7.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/person.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/person.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/person_TjahjonoDGondhowiardjo.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/person_TjahjonoDGondhowiardjo.jpg
--------------------------------------------------------------------------------
/PracticePytorch/faces/personalpic.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/faces/personalpic.jpg
--------------------------------------------------------------------------------
/PracticePytorch/runs/Jun18_14-46-04_Sherlocks-MacBook-Pro.local/events.out.tfevents.1687117564.Sherlocks-MacBook-Pro.local.51259.0:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/runs/Jun18_14-46-04_Sherlocks-MacBook-Pro.local/events.out.tfevents.1687117564.Sherlocks-MacBook-Pro.local.51259.0
--------------------------------------------------------------------------------
/PracticePytorch/runs/Jun18_15-03-00_Sherlocks-MacBook-Pro.local/events.out.tfevents.1687118580.Sherlocks-MacBook-Pro.local.51259.1:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/runs/Jun18_15-03-00_Sherlocks-MacBook-Pro.local/events.out.tfevents.1687118580.Sherlocks-MacBook-Pro.local.51259.1
--------------------------------------------------------------------------------
/PracticePytorch/runs/Jun18_15-03-00_Sherlocks-MacBook-Pro.local/events.out.tfevents.1687119446.Sherlocks-MacBook-Pro.local.51259.2:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/runs/Jun18_15-03-00_Sherlocks-MacBook-Pro.local/events.out.tfevents.1687119446.Sherlocks-MacBook-Pro.local.51259.2
--------------------------------------------------------------------------------
/PracticePytorch/runs/efficientnet_b0_pizza_steak_sushi/events.out.tfevents.1686965428.Sherlocks-MacBook-Pro.local.31548.0:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/runs/efficientnet_b0_pizza_steak_sushi/events.out.tfevents.1686965428.Sherlocks-MacBook-Pro.local.31548.0
--------------------------------------------------------------------------------
/PracticePytorch/runs/efficientnet_b0_pizza_steak_sushi/events.out.tfevents.1686965887.Sherlocks-MacBook-Pro.local.32194.0:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/runs/efficientnet_b0_pizza_steak_sushi/events.out.tfevents.1686965887.Sherlocks-MacBook-Pro.local.32194.0
--------------------------------------------------------------------------------
/PracticePytorch/runs/efficientnet_b0_pizza_steak_sushi/events.out.tfevents.1686966251.Sherlocks-MacBook-Pro.local.32194.1:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/PracticePytorch/runs/efficientnet_b0_pizza_steak_sushi/events.out.tfevents.1686966251.Sherlocks-MacBook-Pro.local.32194.1
--------------------------------------------------------------------------------
/Python&Probability&Statistics/BasicPython/Numpy_Exercise.ipynb:
--------------------------------------------------------------------------------
1 | {
2 | "cells": [
3 | {
4 | "cell_type": "code",
5 | "execution_count": 1,
6 | "metadata": {
7 | "collapsed": true,
8 | "ExecuteTime": {
9 | "end_time": "2023-06-30T20:46:19.620108Z",
10 | "start_time": "2023-06-30T20:46:19.543782Z"
11 | }
12 | },
13 | "outputs": [
14 | {
15 | "data": {
16 | "text/plain": "(array([1., 2., 3.], dtype=float32), 4)"
17 | },
18 | "execution_count": 1,
19 | "metadata": {},
20 | "output_type": "execute_result"
21 | }
22 | ],
23 | "source": [
24 | "import numpy as np\n",
25 | "\n",
26 | "x = np.array([1, 2, 3], dtype = np.float32) # 1 byte = 8 bits, 4 bytes = 32 bits\n",
27 | "\n",
28 | "x, x.itemsize"
29 | ]
30 | },
31 | {
32 | "cell_type": "code",
33 | "execution_count": 2,
34 | "outputs": [
35 | {
36 | "data": {
37 | "text/plain": "(array([0.841471 , 0.9092974, 0.14112 ], dtype=float32),\n array([0.7615942, 0.9640276, 0.9950548], dtype=float32))"
38 | },
39 | "execution_count": 2,
40 | "metadata": {},
41 | "output_type": "execute_result"
42 | }
43 | ],
44 | "source": [
45 | "np.sin(x), np.tanh(x)"
46 | ],
47 | "metadata": {
48 | "collapsed": false,
49 | "ExecuteTime": {
50 | "end_time": "2023-06-30T20:48:43.672557Z",
51 | "start_time": "2023-06-30T20:48:43.668485Z"
52 | }
53 | }
54 | },
55 | {
56 | "cell_type": "code",
57 | "execution_count": 6,
58 | "outputs": [
59 | {
60 | "data": {
61 | "text/plain": "(array([[1., 1., 1., 1.],\n [1., 1., 1., 1.],\n [1., 1., 1., 1.]]),\n array([[1., 1.],\n [1., 9.]]))"
62 | },
63 | "execution_count": 6,
64 | "metadata": {},
65 | "output_type": "execute_result"
66 | }
67 | ],
68 | "source": [
69 | "x = np.ones((3, 3))\n",
70 | "# add one dim\n",
71 | "y = x[:, [0, 1, 2, 2]] # then y has independent memory area\n",
72 | "\n",
73 | "# however, if z is generated by slicing x\n",
74 | "z = x[:2, :2] # then z is just a view of x, so change the elements in x also change the elements in z\n",
75 | "\n",
76 | "x[1, 1] = 9\n",
77 | "\n",
78 | "y, z"
79 | ],
80 | "metadata": {
81 | "collapsed": false,
82 | "ExecuteTime": {
83 | "end_time": "2023-06-30T23:25:25.796517Z",
84 | "start_time": "2023-06-30T23:25:25.793683Z"
85 | }
86 | }
87 | },
88 | {
89 | "cell_type": "code",
90 | "execution_count": 13,
91 | "outputs": [
92 | {
93 | "data": {
94 | "text/plain": "(array([[1., 1., 1.],\n [1., 9., 1.],\n [1., 1., 1.]]),\n array([[1., 1.],\n [1., 1.]]))"
95 | },
96 | "execution_count": 13,
97 | "metadata": {},
98 | "output_type": "execute_result"
99 | }
100 | ],
101 | "source": [
102 | "# Explicit force a copy\n",
103 | "x = np.ones((3, 3))\n",
104 | "z = x[:2, :2].copy() # z becomes independent\n",
105 | "\n",
106 | "x[1, 1] = 9\n",
107 | "\n",
108 | "x, z"
109 | ],
110 | "metadata": {
111 | "collapsed": false,
112 | "ExecuteTime": {
113 | "end_time": "2023-07-01T00:15:32.455736Z",
114 | "start_time": "2023-07-01T00:15:32.452127Z"
115 | }
116 | }
117 | },
118 | {
119 | "cell_type": "code",
120 | "execution_count": 14,
121 | "outputs": [
122 | {
123 | "data": {
124 | "text/plain": "( C_CONTIGUOUS : True\n F_CONTIGUOUS : False\n OWNDATA : True\n WRITEABLE : True\n ALIGNED : True\n WRITEBACKIFCOPY : False,\n '\\n',\n C_CONTIGUOUS : True\n F_CONTIGUOUS : False\n OWNDATA : True\n WRITEABLE : True\n ALIGNED : True\n WRITEBACKIFCOPY : False)"
125 | },
126 | "execution_count": 14,
127 | "metadata": {},
128 | "output_type": "execute_result"
129 | }
130 | ],
131 | "source": [
132 | "x.flags, \"\\n\", z.flags # \"flags\" can help"
133 | ],
134 | "metadata": {
135 | "collapsed": false,
136 | "ExecuteTime": {
137 | "end_time": "2023-07-01T00:15:32.783013Z",
138 | "start_time": "2023-07-01T00:15:32.777025Z"
139 | }
140 | }
141 | },
142 | {
143 | "cell_type": "code",
144 | "execution_count": null,
145 | "outputs": [],
146 | "source": [],
147 | "metadata": {
148 | "collapsed": false
149 | }
150 | }
151 | ],
152 | "metadata": {
153 | "kernelspec": {
154 | "display_name": "Python 3",
155 | "language": "python",
156 | "name": "python3"
157 | },
158 | "language_info": {
159 | "codemirror_mode": {
160 | "name": "ipython",
161 | "version": 2
162 | },
163 | "file_extension": ".py",
164 | "mimetype": "text/x-python",
165 | "name": "python",
166 | "nbconvert_exporter": "python",
167 | "pygments_lexer": "ipython2",
168 | "version": "2.7.6"
169 | }
170 | },
171 | "nbformat": 4,
172 | "nbformat_minor": 0
173 | }
174 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Paper Replication Repo
2 | - This repo includes detailed replications of popular computer vision papers with exhaustive
3 | annotations.
4 | - All the codes in this project is written by Pytorch.
5 |
6 | - New to Pytorch? [See Your First Pytorch Tutor](https://pytorch.org/tutorials/)
7 | - [The second best place to learn Pytorch](https://www.learnpytorch.io/)
8 | - Important architectures and mechanisms in Machine Learning and Deep Learning are included and
9 | well annotated.
10 |
11 |
12 |
13 |
14 | - [x] Last Update Time: June 19 2023
15 | - [ ] In Process: Swin Transformer from Scratch
16 |
17 | ---
18 | ## Content
19 |
20 | - [Self-Attention & Multi-Head Self-Attention Mechanism and Implementation from Scratch](https://github.com/PeiranLi0930/TorchProject/blob/main/PaperReplicate/Self_Attention_from_Scratch/Self-Attention%20and%20Multi-head%20Attention%20Mechanism%20036331bdfc7649238f86306bb44bed38.md)
21 |
22 | - [Implementation Notebook](https://github.com/PeiranLi0930/TorchProject/blob/main/PaperReplicate/Self_Attention_from_Scratch/self_attention_mechanism.ipynb)
23 | - [Functionalized Version](https://github.com/PeiranLi0930/TorchProject/blob/main/PaperReplicate/Self_Attention_from_Scratch/self_attention.py)
24 |
25 |
26 |
27 | ---
28 | ### Appreciations
29 |
30 | - The [Holy Land](https://www.wisc.edu/) giving me best working environments
31 | - My first Pytorch tutor: [Daniel Bourke](https://github.com/mrdbourke), the creator of [Zero to
32 | Mastery Learn PyTorch for Deep Learning](https://www.learnpytorch.io/)
33 | - Significant Guiders on my path to Machine Learning and Deep Learning: [Robert Nowak](https://nowak.ece.wisc.edu/)
34 | , [Andrew Ng](https://www.andrewng.org/), [Feifei Li](https://profiles.stanford.edu/fei-fei-li)
35 | , [Sharon Zhou](https://sharonzhou.me/), [Sebastian Raschka](https://sebastianraschka.com/)
36 | - I wish to express my boundless gratitude to Professor Vikas Singh. I appreciate him enduring my incessant ignorance, offering me ample patience and tolerance. He is a vital guide on my academic journey.
37 |
38 |
39 |
40 | © 2023 - Peiran in USA - All rights reserved
41 |
42 | ---
--------------------------------------------------------------------------------
/Signal_Processing/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/.DS_Store
--------------------------------------------------------------------------------
/Signal_Processing/Intro/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/Intro/.DS_Store
--------------------------------------------------------------------------------
/Signal_Processing/Intro/glassDance.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/Intro/glassDance.mat
--------------------------------------------------------------------------------
/Signal_Processing/Intro/sigprocMXC_filterGlass.m:
--------------------------------------------------------------------------------
1 | %%
2 | % COURSE: Signal processing problems, solved in MATLAB and Python
3 | % SECTION: Introduction
4 | % VIDEO: Having fun with filtered Glass dance
5 | % Instructor: sincxpress.com
6 | %
7 | %%
8 |
9 | % load the file
10 | load glassDance.mat
11 | % this is a clip of Philip Glass, Dance VII (https://www.youtube.com/watch?v=LpewOlR-z_4)
12 |
13 |
14 | % play the music!
15 | soundsc(glassclip,srate)
16 |
17 | % some variables for convenience
18 | pnts = length(glassclip);
19 | timevec = (0:pnts-1)/srate;
20 |
21 | % draw the time-domain signals
22 | figure(1), clf
23 | subplot(511)
24 | plot(timevec,glassclip)
25 | xlabel('Time (s)')
26 |
27 |
28 | %% static power spectrum and pick a frequency range
29 |
30 | % inspect the power spectrum
31 | hz = linspace(0,srate/2,floor(length(glassclip)/2)+1);
32 | powr = abs(fft(glassclip(:,1))/pnts);
33 |
34 | subplot(512), cla
35 | plot(hz,powr(1:length(hz)))
36 | set(gca,'xlim',[100 2000],'ylim',[0 max(powr)])
37 | xlabel('Frequency (Hz)'), ylabel('Amplitude')
38 |
39 |
40 | % pick frequencies to filter
41 | frange = [ 300 460 ];
42 | frange = [ 1000 1100 ];
43 | % frange = [ 1200 1450 ];
44 |
45 |
46 | % design an FIR1 filter
47 | fkern = fir1(2001,frange/(srate/2),'bandpass');
48 |
49 | % apply the filter to the signal
50 | filtglass(:,1) = filtfilt(fkern,1,glassclip(:,1));
51 | filtglass(:,2) = filtfilt(fkern,1,glassclip(:,2));
52 |
53 | % plot the filtered signal power spectrum
54 | hold on
55 | powr = abs(fft(filtglass(:,1))/pnts);
56 | plot(hz,powr(1:length(hz)),'r')
57 |
58 |
59 | % plot the time-frequency response
60 | subplot(5,1,3:5)
61 | spectrogram(glassclip(:,1),hann(round(srate/10)),[],[],srate,'yaxis');
62 | hold on
63 | plot(timevec([1 1; end end]),frange([1 2; 1 2])/1000,'k:','linew',2)
64 |
65 | % NOTE: use the following line in Octave
66 | %[powspect,frex,time] = specgram(glassclip(:,1),1000,srate,hann(round(srate/10)));
67 |
68 |
69 | set(gca,'ylim',[0 2])
70 |
71 | %% play the sound
72 |
73 | soundsc(filtglass,srate)
74 |
75 | %% done.
76 |
77 |
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/__pycache__/slective_search.cpython-310.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/__pycache__/slective_search.cpython-310.pyc
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/2012_001297.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/2012_001297.jpg
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/cats.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/cats.png
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/example_id1.JPG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/example_id1.JPG
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/example_id2.JPG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/example_id2.JPG
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/example_id3.JPG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/example_id3.JPG
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/example_id4.JPG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/example_id4.JPG
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/image/example_image_easy.JPG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/SelectiveSearch/image/example_image_easy.JPG
--------------------------------------------------------------------------------
/Signal_Processing/SelectiveSearch/slective_search.py:
--------------------------------------------------------------------------------
1 | import pathlib
2 |
3 | import matplotlib.pyplot as plt
4 | import skimage
5 | import os
6 | import numpy as np
7 | import seaborn as sns
8 | import pandas as pd
9 | import scipy.misc
10 | import skimage.segmentation
11 | import skimage.feature
12 | from copy import copy
13 | import cv2
14 |
15 | def image_seg(img_8bit, scale = 1.0, sigma = 0.8, min_size = 50):
16 | """
17 | Generate image segmentations and mask that to original image
18 | """
19 |
20 | img_float = skimage.util.img_as_float(img_8bit)
21 | img_mask = skimage.segmentation.felzenszwalb(img_8bit,
22 | scale = scale,
23 | sigma = sigma,
24 | min_size = min_size)
25 | img = np.dstack([img_8bit, img_mask]) # the img[:, :, 3] is the segmentation mask
26 |
27 | return(img)
28 |
29 | def extract_region(img):
30 | """
31 | For each segmented region, extract the smallest rectangle regions covering the smallest region.
32 | """
33 | img_segment = img[:, :, 3]
34 | region = {}
35 |
36 | for y, i in enumerate(img_segment):
37 | for x, j in enumerate(i):
38 |
39 | if j not in region:
40 | region[j] = {"up_left_x" : np.Inf,
41 | "up_left_y" : np.Inf,
42 | "down_right_x" : 0,
43 | "down_right_y" : 0,
44 | "region" : j}
45 |
46 | if region[j]["up_left_x"] > x:
47 | region[j]["up_left_x"] = x
48 | if region[j]["up_left_y"] > y:
49 | region[j]["up_left_y"] = y
50 | if region[j]["down_right_x"] < x:
51 | region[j]["down_right_x"] = x
52 | if region[j]["down_right_y"] < y:
53 | region[j]["down_right_y"] = y
54 |
55 | copied_region_dict = copy(region)
56 |
57 | for key in region.keys():
58 | if (region[key]["down_right_x"] == region[key]["up_left_x"] or
59 | region[key]["down_right_y"] == region[key]["up_left_y"]):
60 | del copied_region_dict[key]
61 |
62 | return copied_region_dict
63 |
64 | def plt_rectangle(plt, label, x1, y1, x2, y2, color = "yellow", alpha = 0.5):
65 | linewidth = 3
66 | if type(label) == list:
67 | linewidth = len(label) * 3 + 2
68 | label = ""
69 |
70 | plt.text(x1, y1, label, fontsize = 20, backgroundcolor = color, alpha = alpha)
71 | plt.plot([x1, x1], [y1, y2], linewidth = linewidth, color = color, alpha = alpha)
72 | plt.plot([x2, x2], [y1, y2], linewidth = linewidth, color = color, alpha = alpha)
73 | plt.plot([x1, x2], [y1, y1], linewidth = linewidth, color = color, alpha = alpha)
74 | plt.plot([x1, x2], [y2, y2], linewidth = linewidth, color = color, alpha = alpha)
75 |
76 | def calc_texture_gradient(img):
77 | """
78 | Calculate texture gradient.
79 | The original Selective Search algo used Gaussian Derivative for 8 orientation.
80 | Here, we use LBP.
81 | """
82 | ret = np.zeros(img.shape[:3])
83 | for c in (0, 1, 2):
84 | ret[:, :, c] = skimage.feature.local_binary_pattern(img[:, :, c], 8, 1.0)
85 |
86 | return ret
87 |
88 | def to_hsv(img):
89 | """
90 | IMG from RGB to HSV. (Hue, Saturation, Value)
91 | """
92 | hsv = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
93 | return hsv
94 |
95 | def generate_hist(img, minhist = 0, maxhist = 1):
96 | """
97 | calculate colour histogram for each region
98 |
99 | the size of output histogram will be BINS * COLOUR_CHANNELS(3)
100 |
101 | number of bins is 25 as same as [uijlings_ijcv2013_draft.pdf]
102 |
103 | extract HSV
104 |
105 | len(hist) = BINS * 3
106 | hist[:BINS] = [0, 10, 20, 0,...,0] meaning that
107 | there are 10 pixels that have values between (maxhist - minhist)/BINS*1 and (maxhist - minhist)/BINS*2
108 | there are 20 pixels that have values between (maxhist - minhist)/BINS*2 and (maxhist - minhist)/BINS*3
109 |
110 | """
111 |
112 | BINS = 25 # length of unit hist
113 | hist = np.array([])
114 | for channel in range(3):
115 | c = img[:, :, channel]
116 | # hist will return back a tuple (histogram, bin_edges)
117 | # e.g. (array([2037, 2002, 2016, 2041, 1972, 2023, 2022, 1994, 2010, 2047, 2073,
118 | # 1933, 2088, 1976, 2059, 1986, 1980, 2021, 2028, 1990, 1984, 1914,
119 | # 1936, 2077, 1967]),
120 | # array([0. , 0.04, 0.08, 0.12, 0.16, 0.2 , 0.24, 0.28, 0.32, 0.36, 0.4 ,
121 | # 0.44, 0.48, 0.52, 0.56, 0.6 , 0.64, 0.68, 0.72, 0.76, 0.8 , 0.84,
122 | # 0.88, 0.92, 0.96, 1. ], dtype=float32))
123 | hist = np.concatenate([hist] + [np.histogram(c, BINS, (minhist, maxhist))[0]])
124 |
125 | hist = hist / len(img) # normalize
126 | return hist
127 |
128 | def augmented_regions_with_histogram_info(texture_grad, img, region_dict: dict, hsv, tex_trad):
129 | for k, v in list(region_dict.items()):
130 | masked_pixels = hsv[img[:, :, 3] == k]
131 | region_dict[k]["size"] = len(masked_pixels / 4) # 4 channels
132 | region_dict[k]["hist_channel"] = generate_hist(masked_pixels, minhist = 0, maxhist = 1)
133 | region_dict[k]["hist_texture"] = generate_hist(texture_grad[img[:, :, 3] == k], minhist =
134 | 0, maxhist = 2 ** 8 - 1)
135 |
136 | return region_dict
137 |
138 | def extract_neighbours(regions):
139 | def intersect(a, b) -> bool:
140 | """
141 | Determine whether there are intersection between two windows
142 | """
143 | if (a["up_left_x"] < b["up_left_x"] < a["down_right_x"] and a["up_left_y"] < b[
144 | "up_left_y"] < a["down_right_y"]) or \
145 | (a["up_left_x"] < b["down_right_x"] < a["down_right_x"] and a["up_left_y"] < b[
146 | "up_left_y"] < a["down_right_y"]) or \
147 | (a["up_left_x"] < b["down_right_x"] < a["down_right_x"] and a["up_left_y"] < b[
148 | "down_right_y"] < a["down_right_y"]) or \
149 | (a["up_left_x"] < b["up_left_x"] < a["down_right_x"] and a["up_left_y"] < b[
150 | "down_right_y"] < a["down_right_y"]):
151 | return True
152 | return False
153 |
154 | region_dict_list = list(regions.items()) # [("": ), ("": ), ("": ), ...]
155 | neighbors = []
156 |
157 | for current, a in enumerate(region_dict_list[:-1]):
158 | for b in region_dict_list[current + 1:]:
159 | if intersect(a[1], b[1]):
160 | neighbors.append((a, b))
161 |
162 | return neighbors
163 |
164 |
165 | if __name__ == '__main__':
166 | img = cv2.imread(os.path.join("image", "example_id1.JPG"), cv2.IMREAD_COLOR)
167 |
168 | segmented_img = image_seg(img)
169 | R = extract_region(segmented_img)
170 |
171 | figsize = (20, 20)
172 | plt.figure(figsize = figsize)
173 | plt.imshow(img[:, :, :3])
174 | for item, color in zip(R.values(), sns.xkcd_rgb.values()):
175 | x1 = item["up_left_x"]
176 | y1 = item["up_left_y"]
177 | x2 = item["down_right_x"]
178 | y2 = item["down_right_y"]
179 | label = item["region"]
180 | plt_rectangle(plt, label, x1, y1, x2, y2, color = color)
181 | plt.show()
182 |
183 | # plt.figure(figsize = figsize)
184 | # plt.imshow(img[:, :, 3])
185 | # for item, color in zip(R.values(), sns.xkcd_rgb.values()):
186 | # x1 = item["min_x"]
187 | # y1 = item["min_y"]
188 | # x2 = item["max_x"]
189 | # y2 = item["max_y"]
190 | # label = item["labels"][0]
191 | # plt_rectangle(plt, label, x1, y1, x2, y2, color = color)
192 | # plt.show()
193 |
194 |
195 | # np.random.seed(4)
196 | # list_path = os.listdir("./image")
197 | # total_num = len(list_path)
198 | # rand_img_path = np.random.choice(list_path, 1)
199 | # img_8bit = cv2.imread(os.path.join("image", rand_img_path[0]), cv2.IMREAD_COLOR)
200 | #
201 | # plt.imshow(to_hsv(img_8bit))
202 | # plt.show()
203 |
204 | # img = img_8bit[:, :, :3]
205 | # img = calc_texture_gradient(img)
206 | # plt.imshow(img)
207 | # plt.show()
208 |
209 | # for img in rand_img_path:
210 | # img_8bit = cv2.imread(os.path.join("image", img), cv2.IMREAD_COLOR)
211 | # seged_img = image_seg(img_8bit, *Config)
212 | #
213 | # fig = plt.figure(figsize = (15, 30))
214 | #
215 | # ax = fig.add_subplot(1, 2, 1)
216 | # ax.imshow(img_8bit)
217 | # ax.set_title("original image")
218 | # ax = fig.add_subplot(1, 2, 2)
219 | # ax.imshow(seged_img[:, :, 3]) # the segmentation output
220 | # ax.set_title("skimage.segmentation.felzenszwalb, N unique region = {}".format(
221 | # len(np.unique(seged_img[:, :, 3]))))
222 | #
223 | # plt.show()
224 |
225 | # Region Extraction
226 |
227 |
228 |
229 |
230 |
231 |
232 |
233 |
--------------------------------------------------------------------------------
/Signal_Processing/TimeSeriesDenoising/.DS_Store:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/TimeSeriesDenoising/.DS_Store
--------------------------------------------------------------------------------
/Signal_Processing/TimeSeriesDenoising/denoising_codeChallenge.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/TimeSeriesDenoising/denoising_codeChallenge.mat
--------------------------------------------------------------------------------
/Signal_Processing/TimeSeriesDenoising/emg4TKEO.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/TimeSeriesDenoising/emg4TKEO.mat
--------------------------------------------------------------------------------
/Signal_Processing/TimeSeriesDenoising/eyedat.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/TimeSeriesDenoising/eyedat.mat
--------------------------------------------------------------------------------
/Signal_Processing/TimeSeriesDenoising/read_data.py:
--------------------------------------------------------------------------------
1 | # 导入所需的库
2 | import scipy.io as sio
3 |
4 | # 使用loadmat函数读取.mat文件
5 | data = sio.loadmat('Signal_Processing/TimeSeriesDenoising/templateProjection.mat')
6 |
7 |
--------------------------------------------------------------------------------
/Signal_Processing/TimeSeriesDenoising/templateProjection.mat:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/Signal_Processing/TimeSeriesDenoising/templateProjection.mat
--------------------------------------------------------------------------------
/__pycache__/data_setup.cpython-310.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/data_setup.cpython-310.pyc
--------------------------------------------------------------------------------
/__pycache__/data_setup.cpython-38.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/data_setup.cpython-38.pyc
--------------------------------------------------------------------------------
/__pycache__/helper_functions.cpython-38.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/helper_functions.cpython-38.pyc
--------------------------------------------------------------------------------
/__pycache__/helper_functions.cpython-39.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/helper_functions.cpython-39.pyc
--------------------------------------------------------------------------------
/__pycache__/module.cpython-38.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/module.cpython-38.pyc
--------------------------------------------------------------------------------
/__pycache__/utils.cpython-310.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/utils.cpython-310.pyc
--------------------------------------------------------------------------------
/__pycache__/utils.cpython-38.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/PeiranLi0930/TorchProject/d376d4f8a78c4ca45315964ab89ab003d3b81378/__pycache__/utils.cpython-38.pyc
--------------------------------------------------------------------------------