├── .github
└── workflows
│ └── check-pull-request.yaml
├── .gitignore
├── LICENCE.md
├── README.md
├── docs
├── pdf-screenshot.png
├── terminology.md
└── video-screenshot.png
├── fonts
└── DejaVuSansCondensed.ttf
├── requirements.txt
├── src
├── __init__.py
├── content_segment_exporter.py
├── main.py
├── plot.py
├── subtitle_part.py
├── subtitle_segment_finder.py
├── subtitle_srt_parser.py
├── subtitle_webvtt_parser.py
├── time_utils.py
└── video_segment_finder.py
└── tests
├── __init__.py
├── snapshots
├── __init__.py
├── snap_test_main
│ ├── video_1_with_srt_file.pdf
│ ├── video_1_with_webvtt_file.pdf
│ ├── video_1_without_subtitles.pdf
│ ├── video_2_with_srt_file.pdf
│ ├── video_2_with_webvtt_file.pdf
│ ├── video_2_without_subtitles.pdf
│ ├── video_7_with_srt_file.pdf
│ ├── video_7_without_subtitles.pdf
│ ├── video_8_with_srt_file.pdf
│ ├── video_8_with_webvtt_file.pdf
│ └── video_8_without_subtitles.pdf
└── snap_test_subtitle_segment_finder.py
├── subtitles
├── subtitles_1.srt
├── subtitles_1.vtt
├── subtitles_2.srt
├── subtitles_2.vtt
├── subtitles_7.srt
├── subtitles_8.srt
└── subtitles_8.vtt
├── test_main.py
├── test_subtitle_segment_finder.py
├── test_subtitle_srt_parser.py
├── test_subtitle_webvtt_parser.py
├── test_time_utils.py
├── test_video_segment_finder.py
├── utils
├── __init__.py
└── pdf_snapshot.py
└── videos
├── input_1.mp4
├── input_2.mp4
├── input_3.mp4
├── input_4.mp4
├── input_5.mp4
├── input_6.mp4
├── input_7.mp4
└── input_8.mp4
/.github/workflows/check-pull-request.yaml:
--------------------------------------------------------------------------------
1 | ##################################################################
2 | # A Github action used to check pull requests
3 | # More info at https://docs.github.com/en/actions/learn-github-actions/workflow-syntax-for-github-actions
4 | ##################################################################
5 |
6 | name: Check the pull request automatically
7 | on:
8 | push:
9 | branches: [main]
10 | pull_request:
11 | branches: [main]
12 |
13 | jobs:
14 | run-tests:
15 | name: Runs all tests
16 | runs-on: ubuntu-latest
17 | steps:
18 | - name: Checkout branch
19 | uses: actions/checkout@v2
20 |
21 | - name: Setup Python
22 | uses: actions/setup-python@v4
23 | with:
24 | python-version: "3.9"
25 |
26 | - name: Cache dependencies
27 | uses: actions/cache@v3.0.11
28 | with:
29 | path: "**/node_modules"
30 | key: ${{ runner.os }}-build-${{ env.cache-name }}-${{ hashFiles('**/requirements.txt') }}
31 |
32 | - name: Install dependencies
33 | run: |
34 | sudo apt install ghostscript
35 | sudo apt install imagemagick
36 | sudo apt install graphicsmagick
37 | sudo apt install pdftk
38 |
39 | # Remove the policy to allow pdf generation
40 | sudo rm /etc/ImageMagick-6/policy.xml
41 |
42 | python -m pip install --upgrade pip
43 | pip install -r requirements.txt
44 |
45 | - name: Run all test cases
46 | run: |
47 | python3 -m unittest discover
48 |
49 | run-app:
50 | name: Run the app
51 | runs-on: ubuntu-latest
52 | steps:
53 | - name: Checkout branch
54 | uses: actions/checkout@v2
55 |
56 | - name: Setup Python
57 | uses: actions/setup-python@v4
58 | with:
59 | python-version: "3.9"
60 |
61 | - name: Cache dependencies
62 | uses: actions/cache@v3.0.11
63 | with:
64 | path: "**/node_modules"
65 | key: ${{ runner.os }}-build-${{ env.cache-name }}-${{ hashFiles('**/requirements.txt') }}
66 |
67 | - name: Install dependencies
68 | run: |
69 | python -m pip install --upgrade pip
70 | pip install -r requirements.txt
71 |
72 | - name: Run the app with sample input
73 | run: python3 -m src.main tests/videos/input_1.mp4 -s tests/subtitles/subtitles_1.vtt -o output.pdf
74 |
--------------------------------------------------------------------------------
/.gitignore:
--------------------------------------------------------------------------------
1 | # Byte-compiled / optimized / DLL files
2 | __pycache__/
3 | *.py[cod]
4 | *$py.class
5 | *.pkl
6 |
7 | # C extensions
8 | *.so
9 |
10 | # Distribution / packaging
11 | .Python
12 | build/
13 | develop-eggs/
14 | dist/
15 | downloads/
16 | eggs/
17 | .eggs/
18 | lib/
19 | lib64/
20 | parts/
21 | sdist/
22 | var/
23 | wheels/
24 | share/python-wheels/
25 | *.egg-info/
26 | .installed.cfg
27 | *.egg
28 | MANIFEST
29 |
30 | # PyInstaller
31 | # Usually these files are written by a python script from a template
32 | # before PyInstaller builds the exe, so as to inject date/other infos into it.
33 | *.manifest
34 | *.spec
35 |
36 | # Installer logs
37 | pip-log.txt
38 | pip-delete-this-directory.txt
39 |
40 | # Unit test / coverage reports
41 | htmlcov/
42 | .tox/
43 | .nox/
44 | .coverage
45 | .coverage.*
46 | .cache
47 | nosetests.xml
48 | coverage.xml
49 | *.cover
50 | *.py,cover
51 | .hypothesis/
52 | .pytest_cache/
53 | cover/
54 |
55 | # Translations
56 | *.mo
57 | *.pot
58 |
59 | # Django stuff:
60 | *.log
61 | local_settings.py
62 | db.sqlite3
63 | db.sqlite3-journal
64 |
65 | # Flask stuff:
66 | instance/
67 | .webassets-cache
68 |
69 | # Scrapy stuff:
70 | .scrapy
71 |
72 | # Sphinx documentation
73 | docs/_build/
74 |
75 | # PyBuilder
76 | .pybuilder/
77 | target/
78 |
79 | # Jupyter Notebook
80 | .ipynb_checkpoints
81 |
82 | # IPython
83 | profile_default/
84 | ipython_config.py
85 |
86 | # pyenv
87 | # For a library or package, you might want to ignore these files since the code is
88 | # intended to run in multiple environments; otherwise, check them in:
89 | # .python-version
90 |
91 | # pipenv
92 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
93 | # However, in case of collaboration, if having platform-specific dependencies or dependencies
94 | # having no cross-platform support, pipenv may install dependencies that don't work, or not
95 | # install all needed dependencies.
96 | #Pipfile.lock
97 |
98 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow
99 | __pypackages__/
100 |
101 | # Celery stuff
102 | celerybeat-schedule
103 | celerybeat.pid
104 |
105 | # SageMath parsed files
106 | *.sage.py
107 |
108 | # Environments
109 | .env
110 | .venv
111 | env/
112 | venv/
113 | ENV/
114 | env.bak/
115 | venv.bak/
116 | bin/
117 | include/
118 | share/
119 | pyvenv.cfg
120 |
121 | # Spyder project settings
122 | .spyderproject
123 | .spyproject
124 |
125 | # Rope project settings
126 | .ropeproject
127 |
128 | # mkdocs documentation
129 | /site
130 |
131 | # mypy
132 | .mypy_cache/
133 | .dmypy.json
134 | dmypy.json
135 |
136 | # Pyre type checker
137 | .pyre/
138 |
139 | # pytype static type analyzer
140 | .pytype/
141 |
142 | # Cython debug symbols
143 | cython_debug/
144 |
145 | # vs code folder
146 | .vscode/
147 |
148 | # temp folders generated by this project
149 | tmp/
150 | plot-output/
151 | output.pdf
152 |
153 | # Mac files
154 | .DS_Store
--------------------------------------------------------------------------------
/LICENCE.md:
--------------------------------------------------------------------------------
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 | .
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Convert Lecture Videos to PDF
2 |
3 | ### Description
4 |
5 | Want to go through lecture videos faster without missing any information? Wish you can **read** the lecture video instead of watching it? Now you can! With this python app, you can convert lecture videos to PDF files! The PDF file will contain a screenshot of lecture slides presented in the video, along with a transcription of your instructor explaining those lecture slide. It can also handle instructors making annotations on their lecture slides and mild amounts of PowerPoint animations.
6 |
7 | ### Table of Contents
8 |
9 | - Walkthrough
10 | - Getting Started
11 | - Tweeking the Application
12 | - Next steps
13 | - Usage
14 | - Credits
15 | - License
16 |
17 | ### Walkthrough of this project
18 |
19 | Users will need to download a video file of their lecture. For instance, the video file might look like this:
20 |
21 |
22 |
23 |
24 |
25 |
26 |
27 | Users will also need a copy of the video's subtitles.
28 |
29 | After running the command line tool, they will get a PDF that looks like this:
30 |
31 |
32 |
33 |
34 |
35 |
36 |
37 | where each page contains an image of the lecture video, and a transcription of the instructor explaining about that slide.
38 |
39 | ### Getting Started
40 |
41 | 1. Ensure Python3 and Pip is installed on your machine
42 | 2. Next, install package dependencies by running:
43 |
44 | `pip3 install -r requirements.txt`
45 |
46 | 3. Now, run:
47 |
48 | `python3 -m src.main tests/videos/input_1.mp4 -s tests/subtitles/subtitles_1.vtt -o output.pdf`
49 |
50 | to generate a PDF of [this lecture video](tests/videos/input_1.mp4) with [these subtitles](```tests/subtitles/subtitles_1.vtt```)
51 |
52 | Note: If you don't want subtitles in the pdf, you can use the `-S` flag, like:
53 |
54 | `python3 -m src.main tests/videos/input_1.mp4 -S -o output.pdf`
55 |
56 | 4. The generated PDF will be saved as _output.pdf_
57 |
58 | ### Running Tests
59 |
60 | 1. Install graphicsmagick, imagemagick, and pdftk on your machine
61 | 2. To run all unit tests, run `python3 -m unittest discover`
62 | 3. To run a specific unit tests (ex: tests/test_main.py), run `python3 -m unittest tests/test_main.py`
63 |
64 | Note: Running the `tests/test_main.py` takes a while
65 |
66 | ### Tweeking the Application
67 |
68 | This application uses computer vision with OpenCV to detect when the instructor has moved on to the next PowerPoint slide, detect animations, etc.
69 |
70 | You can adjust the sensitivity to video frame changes in the `src/video_segment_finder.py` file. You can also visualize how well the application detect transitions and animations via the `src/plot.py` tool.
71 |
72 | ### Next Steps
73 |
74 | - [ ] Automatically generate subtitles
75 | - [ ] Wrap project into a web app?
76 |
77 | ### Usage
78 |
79 | Please note that this project is used for educational purposes and is not intended to be used commercially. We are not liable for any damages/changes done by this project.
80 |
81 | ### Credits
82 |
83 | Emilio Kartono, who made the entire project.
84 |
85 | The fonts for generating the PDF is from [DejaVu fonts](https://dejavu-fonts.github.io/)
86 |
87 | ### License
88 |
89 | This project is protected under the GNU licence. Please refer to the LICENSE.txt for more information.
90 |
--------------------------------------------------------------------------------
/docs/pdf-screenshot.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/docs/pdf-screenshot.png
--------------------------------------------------------------------------------
/docs/terminology.md:
--------------------------------------------------------------------------------
1 | # Terminology:
2 | * Content segments:
3 | * Is a part of the video from a particular start and end time that is explaining one content
4 | * Ex: 00:02:05 - 00:04:00 of the video is explaining about a slide on planet Mars
5 | * It contains a video segment and its corresponding subtitle segment
6 |
7 | * Video segment:
8 | * Is the last video frame of a content segment
9 | * Ex: the video segment of 00:02:05 - 00:04:00 is the last video frame in 00:02:05 - 00:04:00
10 |
11 | * Subtitle segment:
12 | * Is the text component of a content segment
13 | * Ex: the subtitle segment of 00:02:05 - 00:04:00 is "In this slide, we are going to ..."
14 |
15 | * Subtitle parts:
16 | * Is a subset of the video's subtitles
17 | * Is created when reading in the video's subtitles
--------------------------------------------------------------------------------
/docs/video-screenshot.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/docs/video-screenshot.png
--------------------------------------------------------------------------------
/fonts/DejaVuSansCondensed.ttf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/fonts/DejaVuSansCondensed.ttf
--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | appdirs==1.4.4
2 | black==21.5b2
3 | blis==0.7.5
4 | catalogue==2.0.6
5 | certifi==2020.12.5
6 | chardet==4.0.0
7 | charset-normalizer==2.0.10
8 | click==7.1.2
9 | cymem==2.0.6
10 | distlib==0.3.1
11 | docopt==0.6.2
12 | fastdiff==0.3.0
13 | filelock==3.0.12
14 | fpdf2==2.2.0
15 | greenlet==1.1.1
16 | idna==3.3
17 | Jinja2==3.0.3
18 | langcodes==3.3.0
19 | MarkupSafe==2.0.1
20 | murmurhash==1.0.6
21 | mypy-extensions==0.4.3
22 | numpy==1.19.5
23 | opencv-python==4.5.1.48
24 | packaging==21.3
25 | parameterized==0.8.1
26 | pathspec==0.8.1
27 | pathy==0.6.1
28 | Pillow==8.1.0
29 | pipenv==2020.11.15
30 | playwright==1.15.0
31 | preshed==3.0.6
32 | pydantic==1.8.2
33 | pyee==8.2.2
34 | pyparsing==3.0.6
35 | pypdftk==0.5
36 | pysrt==1.1.2
37 | python-dotenv==0.19.0
38 | regex==2020.11.13
39 | requests==2.27.1
40 | rope==0.18.0
41 | six==1.15.0
42 | smart-open==5.2.1
43 | snapshottest==0.6.0
44 | spacy-legacy==3.0.8
45 | spacy-loggers==1.0.1
46 | srsly==2.4.2
47 | srt==3.5.0
48 | termcolor==1.1.0
49 | thinc==8.0.13
50 | toml==0.10.2
51 | tqdm==4.62.3
52 | typed-ast==1.4.2
53 | typer==0.4.0
54 | typing-extensions==3.7.4.3
55 | urllib3==1.26.8
56 | virtualenv==20.4.4
57 | virtualenv-clone==0.5.4
58 | Wand==0.6.10
59 | wasabi==0.9.0
60 | wasmer==1.1.0
61 | wasmer_compiler_cranelift==1.1.0
62 | websockets==10.0
63 | webvtt-py==0.4.6
64 |
--------------------------------------------------------------------------------
/src/__init__.py:
--------------------------------------------------------------------------------
1 | from .subtitle_part import SubtitlePart
2 | from .subtitle_webvtt_parser import SubtitleWebVTTParser
3 | from .subtitle_segment_finder import SubtitleGenerator, SubtitleSegmentFinder
4 | from .subtitle_srt_parser import SubtitleSRTParser
5 | from .time_utils import (
6 | convert_clock_time_to_timestamp_ms,
7 | convert_timestamp_ms_to_clock_time,
8 | )
9 | from .content_segment_exporter import ContentSegment, ContentSegmentPdfBuilder
10 | from .video_segment_finder import VideoSegmentFinder
11 |
--------------------------------------------------------------------------------
/src/content_segment_exporter.py:
--------------------------------------------------------------------------------
1 | import os
2 | import cv2
3 | from fpdf import FPDF
4 | import tempfile
5 |
6 | from .subtitle_webvtt_parser import SubtitleWebVTTParser
7 | from .subtitle_segment_finder import SubtitleSegmentFinder
8 | from .video_segment_finder import VideoSegmentFinder
9 |
10 |
11 | class ContentSegment:
12 | """This class represents the image and the text that represents one segment of the video
13 | For instance, an image could be a powerpoint slide and the text could be the explaination of that slide
14 |
15 | Attributes
16 | ----------
17 | image : np.array(x, y, 3)
18 | The image
19 | text : str
20 | The text
21 | """
22 |
23 | def __init__(self, image, text):
24 | self.image = image
25 | self.text = text
26 |
27 |
28 | class ContentSegmentPdfBuilder:
29 | """This class creates a PDF from a lecture segment"""
30 |
31 | def generate_pdf(self, pages, output_filepath):
32 | """Generates and saves a PDF from an ordered list of lecture segments
33 |
34 | Parameters
35 | ----------
36 | pages : ContentSegment[]
37 | An ordered list of lecture segments
38 | output_filepath: str
39 | The filepath for the output pdf
40 | """
41 | with tempfile.TemporaryDirectory() as temp_dir_path:
42 | pdf = FPDF()
43 | pdf.add_font("DejaVu", "", "fonts/DejaVuSansCondensed.ttf", uni=True)
44 |
45 | for i in range(0, len(pages)):
46 | # Temporarily save the frames
47 | temp_filepath = os.path.join(temp_dir_path, f"{i}_frame.jpeg")
48 | cv2.imwrite(temp_filepath, pages[i].image)
49 |
50 | pdf.add_page()
51 |
52 | # Add the image
53 | pdf.image(temp_filepath, w=195)
54 |
55 | # Add the captions if exist
56 | if pages[i].text is not None:
57 | pdf.set_font("DejaVu", "", 12)
58 | pdf.multi_cell(0, 10, pages[i].text)
59 |
60 | pdf.output(output_filepath, "F")
61 |
62 |
63 | if __name__ == "__main__":
64 | # Get the selected frames
65 | selected_frames_data = VideoSegmentFinder().get_best_segment_frames(
66 | "../tests/videos/input_1.mp4"
67 | )
68 | frame_nums = sorted(selected_frames_data.keys())
69 | selected_frames = [selected_frames_data[i]["frame"] for i in frame_nums]
70 |
71 | # Get the subtitles for each frame
72 | pager = SubtitleSegmentFinder(
73 | SubtitleWebVTTParser("../tests/subtitles/subtitles_1.vtt").get_subtitle_parts()
74 | )
75 | subtitle_breaks = [selected_frames_data[i]["timestamp"] for i in frame_nums]
76 | pages = pager.get_subtitle_segments(subtitle_breaks)
77 |
78 | # Merge the frame and subtitles for each frame to create a pdf
79 | video_subtitle_pages = []
80 |
81 | for i in range(0, len(selected_frames)):
82 | frame = selected_frames[i]
83 | subtitle_page = pages[i]
84 | video_subtitle_pages.append(ContentSegment(frame, subtitle_page))
85 |
86 | printer = ContentSegmentPdfBuilder()
87 | pdf = printer.generate_pdf(video_subtitle_pages, "myfile.pdf")
88 |
--------------------------------------------------------------------------------
/src/main.py:
--------------------------------------------------------------------------------
1 | import sys
2 | import argparse
3 | from .subtitle_segment_finder import SubtitleGenerator, SubtitleSegmentFinder
4 | from .subtitle_webvtt_parser import SubtitleWebVTTParser
5 | from .subtitle_srt_parser import SubtitleSRTParser
6 | from .video_segment_finder import VideoSegmentFinder
7 | from .content_segment_exporter import ContentSegment, ContentSegmentPdfBuilder
8 |
9 |
10 | class CommandLineArgRunner:
11 | def __init__(self):
12 | self.parser = argparse.ArgumentParser(
13 | description="Generate a readable pdf from lecture videos"
14 | )
15 | self.parser.add_argument("video", type=str, help="File path to lecture video")
16 | self.parser.add_argument(
17 | "-s",
18 | "--subtitle",
19 | type=str,
20 | default=None,
21 | help="File path to video subtitle. If omitted, it will generate subtitles",
22 | )
23 | self.parser.add_argument(
24 | "-S",
25 | "--skip-subtitles",
26 | action="store_true",
27 | help="If flag is set, it will ignore setting subtitles to lecture slides",
28 | )
29 | self.parser.add_argument(
30 | "-o",
31 | "--output",
32 | type=str,
33 | default="output.pdf",
34 | help="Output file to generated pdf",
35 | )
36 |
37 | def run(self, args):
38 | opts = self.parser.parse_args(args)
39 |
40 | video_filepath = opts.video
41 | subtitle_filepath = opts.subtitle
42 | output_filepath = opts.output
43 | is_skip_subtitles = opts.skip_subtitles
44 |
45 | if is_skip_subtitles and subtitle_filepath is not None:
46 | print("Omit the -S / --skip-subtitles flag to add subtitles to pdf")
47 | raise AssertionError()
48 |
49 | video_segment_finder = VideoSegmentFinder()
50 |
51 | if is_skip_subtitles:
52 | self.__generate_pdf_without_subtitles__(
53 | video_segment_finder, video_filepath, output_filepath
54 | )
55 | else:
56 | if subtitle_filepath is None:
57 | subtitle_parser = SubtitleGenerator(video_filepath)
58 | elif subtitle_filepath.endswith(".srt"):
59 | subtitle_parser = SubtitleSRTParser(subtitle_filepath)
60 | else:
61 | subtitle_parser = SubtitleWebVTTParser(subtitle_filepath)
62 |
63 | self.__generate_pdf_with_subtitles__(
64 | video_segment_finder, video_filepath, subtitle_parser, output_filepath
65 | )
66 |
67 | def __generate_pdf_with_subtitles__(
68 | self, video_segment_finder, video_filepath, subtitle_parser, output_filepath
69 | ):
70 | # Get the selected frames
71 | print("Getting selected frames")
72 | selected_frames_data = video_segment_finder.get_best_segment_frames(
73 | video_filepath
74 | )
75 | frame_nums = sorted(selected_frames_data.keys())
76 | selected_frames = [selected_frames_data[i]["frame"] for i in frame_nums]
77 |
78 | print("Number of frames:", len(selected_frames))
79 |
80 | # Get the subtitles for each frame
81 | print("Getting subtitles for each frame")
82 | segment_finder = SubtitleSegmentFinder(subtitle_parser.get_subtitle_parts())
83 | subtitle_breaks = [selected_frames_data[i]["timestamp"] for i in frame_nums]
84 | segments = segment_finder.get_subtitle_segments(subtitle_breaks)
85 |
86 | # Merge the frame and subtitles for each frame to create a pdf
87 | print("Merging frames and subtitles")
88 | video_subtitle_pages = []
89 |
90 | for i in range(0, len(selected_frames)):
91 | frame = selected_frames[i]
92 | subtitle_page = segments[i]
93 | video_subtitle_pages.append(ContentSegment(frame, subtitle_page))
94 |
95 | print("Generating PDF file")
96 | printer = ContentSegmentPdfBuilder()
97 | printer.generate_pdf(video_subtitle_pages, output_filepath)
98 |
99 | def __generate_pdf_without_subtitles__(
100 | self, video_segment_finder, video_filepath, output_filepath
101 | ):
102 | # Get the selected frames
103 | print("Getting selected frames")
104 | selected_frames_data = video_segment_finder.get_best_segment_frames(
105 | video_filepath
106 | )
107 | frame_nums = sorted(selected_frames_data.keys())
108 | selected_frames = [selected_frames_data[i]["frame"] for i in frame_nums]
109 |
110 | print("Number of frames:", len(selected_frames))
111 |
112 | # Generating PDF file
113 | print("Generating PDF file")
114 | video_subtitle_pages = [
115 | ContentSegment(frame, None) for frame in selected_frames
116 | ]
117 | printer = ContentSegmentPdfBuilder()
118 | printer.generate_pdf(video_subtitle_pages, output_filepath)
119 |
120 |
121 | if __name__ == "__main__":
122 | runner = CommandLineArgRunner()
123 | runner.run(sys.argv[1:])
124 |
--------------------------------------------------------------------------------
/src/plot.py:
--------------------------------------------------------------------------------
1 | import glob
2 | import os
3 | import matplotlib.pyplot as plt
4 | import cv2
5 | from .video_segment_finder import VideoSegmentFinder
6 |
7 |
8 | def plot_timestamps_vs_pixel_change(stats):
9 | simple_timestamp = {}
10 |
11 | for frame_num in stats:
12 | timestamp_ms = stats[frame_num]["timestamp"]
13 | timestamp_seconds = int(float(timestamp_ms) / 100)
14 | num_pixels_changed = stats[frame_num]["num_pixels_changed"]
15 |
16 | if timestamp_seconds not in simple_timestamp:
17 | simple_timestamp[timestamp_seconds] = [sys.maxsize, 0, 0, 0]
18 |
19 | simple_timestamp[timestamp_seconds] = [
20 | min(simple_timestamp[timestamp_seconds][0], num_pixels_changed), # min
21 | max(simple_timestamp[timestamp_seconds][1], num_pixels_changed), # max
22 | simple_timestamp[timestamp_seconds][2] + num_pixels_changed, # avg
23 | simple_timestamp[timestamp_seconds][3] + 1, # num points
24 | ]
25 |
26 | # Compute avg
27 | for k in simple_timestamp:
28 | simple_timestamp[k][2] /= simple_timestamp[k][3]
29 |
30 | x_vals = sorted(simple_timestamp.keys())
31 |
32 | plt.plot(x_vals, [simple_timestamp[x][0] for x in x_vals], label="min")
33 | plt.plot(x_vals, [simple_timestamp[x][1] for x in x_vals], label="max")
34 | plt.plot(x_vals, [simple_timestamp[x][2] for x in x_vals], label="avg")
35 |
36 | plt.xlabel("timestamp")
37 | plt.ylabel("score")
38 |
39 | plt.legend()
40 | plt.show()
41 |
42 |
43 | def save_selected_frames(selected_frames):
44 | # Prepare the output folder
45 | if not os.path.exists("./plot-output"):
46 | os.mkdir("plot-output")
47 | for f in glob.glob("./plot-output/*"):
48 | os.remove(f)
49 |
50 | # Save the frames into the output folder
51 | for frame_num in selected_frames:
52 | timestamp = selected_frames[frame_num]["timestamp"]
53 | pixel_changes = selected_frames[frame_num]["num_pixels_changed"]
54 | frame = selected_frames[frame_num]["frame"]
55 | next_frame = selected_frames[frame_num]["next_frame"]
56 | mask = selected_frames[frame_num]["mask"]
57 |
58 | cv2.imwrite(
59 | "plot-output/{}_{}_cur_frame.jpeg".format(timestamp, pixel_changes), frame
60 | )
61 | cv2.imwrite(
62 | "plot-output/{}_{}_next_frame.jpeg".format(timestamp, pixel_changes),
63 | next_frame,
64 | )
65 | cv2.imwrite(
66 | "plot-output/{}_{}_mask_frame.jpeg".format(timestamp, pixel_changes), mask
67 | )
68 |
69 |
70 | if __name__ == "__main__":
71 | selected_frames, stats = VideoSegmentFinder().get_segment_frames_with_stats(
72 | "./tests/videos/input_6.mp4"
73 | )
74 | save_selected_frames(selected_frames)
75 | plot_timestamps_vs_pixel_change(stats)
76 |
--------------------------------------------------------------------------------
/src/subtitle_part.py:
--------------------------------------------------------------------------------
1 | class SubtitlePart:
2 | """A class that represents a part of the entire video's subtitle
3 |
4 | Attributes
5 | ----------
6 | start_time : int
7 | The starting time of this subtitle's part in milliseconds
8 | end_time : int
9 | The end time of this subtitle's part in milliseconds
10 | text : str
11 | The text corresponding to the subtitle's part
12 | """
13 |
14 | def __init__(self, start_time, end_time, text):
15 | self.start_time = start_time
16 | self.end_time = end_time
17 | self.text = text
18 |
19 | def __str__(self):
20 | return "{}-{}".format(self.start_time, self.end_time)
21 |
22 | def __repr__(self):
23 | return self.__str__()
24 |
--------------------------------------------------------------------------------
/src/subtitle_segment_finder.py:
--------------------------------------------------------------------------------
1 | from .subtitle_webvtt_parser import SubtitleWebVTTParser
2 | from .subtitle_srt_parser import SubtitleSRTParser
3 |
4 |
5 | class SubtitleGenerator:
6 | def __init__(self, video_file):
7 | self.video_file = video_file
8 |
9 | def get_segments(self):
10 | pass
11 |
12 |
13 | class SubtitleSegmentFinder:
14 | """This class finds the best subtitle segments from the end times of video segments"""
15 |
16 | def __init__(self, parts):
17 | self.parts = parts
18 |
19 | def get_subtitle_segments(self, video_segment_end_times):
20 | """Returns the subtitles of video segments given the end times of each video segment
21 |
22 | For instance, given times:
23 | [ "00:00:10", "00:00:20", "00:00:30" ]
24 | it will return the subtitles for times:
25 | 00:00:00 - 00:00:10
26 | 00:00:10 - 00:00:20
27 | 00:00:20 - 00:00:30
28 |
29 | Parameters
30 | ----------
31 | video_segment_end_times : int[]
32 | A list of timestamps representing the end times of each video segment
33 |
34 | Returns
35 | -------
36 | segments : str[]
37 | A list of subtitle segments
38 | """
39 | part_positions = []
40 |
41 | for i in range(len(video_segment_end_times)):
42 | time_break = video_segment_end_times[i]
43 |
44 | prev_time_break = 0
45 | if i > 0:
46 | prev_time_break = video_segment_end_times[i - 1]
47 |
48 | next_time_break = float('inf')
49 | if i < len(video_segment_end_times) - 1:
50 | next_time_break = video_segment_end_times[i + 1]
51 |
52 | pos = self.__get_part_position_of_time_break__(time_break, prev_time_break, next_time_break)
53 | part_positions.append(pos)
54 |
55 | start_pos = (0, 0)
56 | segments = []
57 | for end_pos in part_positions:
58 | segment = None
59 |
60 | if start_pos[0] > end_pos[0]:
61 | segment = ""
62 |
63 | elif start_pos[0] == end_pos[0] and start_pos[1] > end_pos[1]:
64 | segment = ""
65 |
66 | elif start_pos[0] == end_pos[0] and start_pos[1] <= end_pos[1]:
67 | segment = self.parts[start_pos[0]].text[start_pos[1] : end_pos[1] + 1]
68 |
69 | elif start_pos[0] < end_pos[0]:
70 | segment = " ".join(
71 | [self.parts[start_pos[0]].text[start_pos[1] :].strip()]
72 | + [self.parts[i].text for i in range(start_pos[0] + 1, end_pos[0])]
73 | + [self.parts[end_pos[0]].text[0 : end_pos[1] + 1].strip()]
74 | )
75 |
76 | segment = segment.strip()
77 |
78 | start_pos = (end_pos[0], end_pos[1] + 1)
79 |
80 | segments.append(segment)
81 |
82 | return segments
83 |
84 | def __get_part_position_of_time_break__(self, time_break, min_time_break, max_time_break):
85 | min_part_idx = self.__find_part__(min_time_break)
86 | max_part_idx = self.__find_part__(max_time_break)
87 | part_index = self.__find_part__(time_break)
88 |
89 | if min_part_idx is None:
90 | min_part_idx = 0
91 |
92 | if max_part_idx is None:
93 | max_part_idx = len(self.parts)
94 |
95 | # If the page_break_time > last fragment's time, then that page needs to capture the entire thing
96 | if time_break >= self.parts[-1].end_time:
97 | return len(self.parts) - 1, len(self.parts[-1].text) - 1
98 |
99 | if part_index is None:
100 | return 0, -1
101 |
102 | part = self.parts[part_index]
103 |
104 | # Get the char index in the fragment equal to the time_break
105 | ratio = (time_break - part.start_time) / (part.end_time - part.start_time)
106 | part_char_index = int(ratio * len(part.text))
107 |
108 | # Find the nearest position of a '.' left or right of 'part_index' and 'part_char_index'
109 | left_part_index = part_index
110 | left_part_char_index = part_char_index
111 | right_part_index = part_index
112 | right_part_char_index = part_char_index
113 |
114 | while left_part_index >= min_part_idx and right_part_index < max_part_idx:
115 | if self.parts[left_part_index].text[left_part_char_index] == ".":
116 | return left_part_index, left_part_char_index
117 |
118 | if self.parts[right_part_index].text[right_part_char_index] == ".":
119 | return right_part_index, right_part_char_index
120 |
121 | left_part_char_index -= 1
122 | right_part_char_index += 1
123 |
124 | if left_part_char_index < 0:
125 | left_part_index -= 1
126 | left_part_char_index = len(self.parts[left_part_index].text) - 1
127 |
128 | if right_part_char_index >= len(self.parts[right_part_index].text):
129 | right_part_index += 1
130 | right_part_char_index = 0
131 |
132 | while left_part_index >= min_part_idx:
133 | if self.parts[left_part_index].text[left_part_char_index] == ".":
134 | return left_part_index, left_part_char_index
135 |
136 | left_part_char_index -= 1
137 |
138 | if left_part_char_index < 0:
139 | left_part_index -= 1
140 | left_part_char_index = len(self.parts[left_part_index].text) - 1
141 |
142 | while right_part_index < max_part_idx:
143 | if self.parts[right_part_index].text[right_part_char_index] == ".":
144 | return right_part_index, right_part_char_index
145 |
146 | right_part_char_index += 1
147 |
148 | if right_part_char_index >= len(self.parts[right_part_index].text):
149 | right_part_index += 1
150 | right_part_char_index = 0
151 |
152 | # Fallback: return the found part index and part char
153 | return part_index, part_char_index
154 |
155 | def __find_part__(self, timestamp_ms):
156 | left = 0
157 | right = len(self.parts) - 1
158 |
159 | while left <= right:
160 | mid = (left + right) // 2
161 | cur_part = self.parts[mid]
162 |
163 | if cur_part.start_time <= timestamp_ms < cur_part.end_time:
164 | return mid
165 | elif timestamp_ms < cur_part.start_time:
166 | right = mid - 1
167 | else:
168 | left = mid + 1
169 |
170 | return None
171 |
172 |
173 | if __name__ == "__main__":
174 |
175 | def test1():
176 | parser = SubtitleWebVTTParser("../tests/subtitles/subtitles_2.vtt")
177 | parts = parser.get_subtitle_parts()
178 | segment_finder = SubtitleSegmentFinder(parts)
179 |
180 | breaks = [14000, 130000, 338000, 478000, 637000, 652000, 654000]
181 |
182 | """ We will have transcriptions at these times:
183 | > 0 - 14000
184 | > 14000 - 130000
185 | > 130000 - 338000
186 | > 338000 - 478000
187 | > 478000 - 637000
188 | > 637000 - 652000
189 | > 652000 - 654000
190 | """
191 | transcript_pages = segment_finder.get_subtitle_segments(breaks)
192 |
193 | print(len(transcript_pages))
194 | for transcript_page in transcript_pages:
195 | print("-----------------------")
196 | print(transcript_page)
197 | print("-----------------------")
198 |
199 | def test2():
200 | parser = SubtitleSRTParser("../tests/subtitles/subtitles_7.srt")
201 | parts = parser.get_subtitle_parts()
202 | print(len(parts))
203 | segment_finder = SubtitleSegmentFinder(parts)
204 |
205 | breaks = [
206 | 10520.0,
207 | 103680.0,
208 | 143360.0,
209 | 773040.0,
210 | 1118240.0,
211 | 1693000.0,
212 | 1704760.0,
213 | ]
214 |
215 | """ We will have transcriptions at these times:
216 | > 0 - 14000
217 | > 14000 - 130000
218 | > 130000 - 338000
219 | > 338000 - 478000
220 | > 478000 - 637000
221 | > 637000 - 652000
222 | > 652000 - 654000
223 | """
224 | transcript_pages = segment_finder.get_subtitle_segments(breaks)
225 |
226 | print(len(transcript_pages))
227 | for transcript_page in transcript_pages:
228 | print("-----------------------")
229 | print(transcript_page)
230 | print("-----------------------")
231 |
232 | test2()
233 |
--------------------------------------------------------------------------------
/src/subtitle_srt_parser.py:
--------------------------------------------------------------------------------
1 | import srt
2 | from .subtitle_part import SubtitlePart
3 |
4 | class SubtitleSRTParser:
5 | """Parses the subtitles and its parts from a .srt file
6 |
7 | Attributes
8 | ----------
9 | input_file : str
10 | The file path to the subtitles
11 | """
12 |
13 | def __init__(self, input_file):
14 | self.input_file = input_file
15 |
16 | def get_subtitle_parts(self):
17 | """Parses and gets the subtitle parts from the subtitle's file
18 | It also expands the subtitles in cases where there are gaps between subtitles
19 |
20 | Returns
21 | -------
22 | parts : SubtitlePart[]
23 | An ordered list of subtitle parts
24 | """
25 | parts = []
26 | with open(self.input_file, mode='r') as f:
27 | for sub in srt.parse(f):
28 | start_time = self.__convert_timedelta_to_ms__(sub.start)
29 | end_time = self.__convert_timedelta_to_ms__(sub.end)
30 | clean_text = self.__filter_text__(sub.content)
31 |
32 | if len(clean_text) == 0:
33 | continue
34 |
35 | parts.append(SubtitlePart(start_time, end_time, clean_text))
36 |
37 | # Extend certain subtitle times to fill in gaps
38 | for i in range(len(parts) - 1):
39 | cur = parts[i]
40 | next = parts[i + 1]
41 |
42 | if cur.end_time != next.start_time:
43 | cur.end_time = next.start_time
44 |
45 | return parts
46 |
47 | def __convert_timedelta_to_ms__(self, timedelta_obj):
48 | return timedelta_obj.total_seconds() * 1000
49 |
50 | def __filter_text__(self, segment_text):
51 | """Takes in the text of a subtitle segment and cleans it"""
52 | return segment_text.replace("\n", " ").strip()
53 |
--------------------------------------------------------------------------------
/src/subtitle_webvtt_parser.py:
--------------------------------------------------------------------------------
1 | import webvtt
2 | from .subtitle_part import SubtitlePart
3 | from .time_utils import convert_clock_time_to_timestamp_ms
4 |
5 | class SubtitleWebVTTParser:
6 | """Parses the subtitles and its parts from a .vtt file
7 |
8 | Attributes
9 | ----------
10 | input_file : str
11 | The file path to the subtitles
12 | """
13 |
14 | def __init__(self, input_file):
15 | self.input_file = input_file
16 |
17 | def get_subtitle_parts(self):
18 | """Parses and gets the subtitle parts from the subtitle's file
19 | It also expands the subtitles in cases where there are gaps between subtitles
20 |
21 | Returns
22 | -------
23 | parts : SubtitlePart[]
24 | An ordered list of subtitle parts
25 | """
26 | parts = []
27 | for caption in webvtt.read(self.input_file):
28 | start_time = convert_clock_time_to_timestamp_ms(caption.start)
29 | end_time = convert_clock_time_to_timestamp_ms(caption.end)
30 | clean_text = self.__filter_text__(caption.text)
31 |
32 | if len(clean_text) == 0:
33 | continue
34 |
35 | parts.append(SubtitlePart(start_time, end_time, clean_text))
36 |
37 | # Extend certain subtitle times to fill in gaps
38 | for i in range(len(parts) - 1):
39 | cur = parts[i]
40 | next = parts[i + 1]
41 |
42 | if cur.end_time != next.start_time:
43 | cur.end_time = next.start_time
44 |
45 | return parts
46 |
47 | def __filter_text__(self, segment_text):
48 | """Takes in the text of a subtitle segment and cleans it"""
49 | return segment_text.replace("\n", " ").strip()
50 |
--------------------------------------------------------------------------------
/src/time_utils.py:
--------------------------------------------------------------------------------
1 | def convert_clock_time_to_timestamp_ms(clock_time):
2 | """Converts time from HH:MM:SS format to timestamp format (in milliseconds)
3 | For instance, given "00:05:38", it will return 338000
4 |
5 | Parameters
6 | ----------
7 | clock_time : str
8 | The clock time in HH:mm:ss format
9 |
10 | Returns
11 | -------
12 | timestamp_ms : int
13 | Timestamp in milliseconds
14 | """
15 | clock_time_parts = clock_time.split(":")
16 | if len(clock_time_parts) != 3:
17 | raise Exception(
18 | "Illegal argument! Expected 3 parts, instead {} in {}".format(
19 | len(clock_time_parts), clock_time
20 | )
21 | )
22 |
23 | hours = int(clock_time_parts[0])
24 | minutes = int(clock_time_parts[1])
25 | seconds = float(clock_time_parts[2])
26 |
27 | return (hours * 3600000) + (minutes * 60000) + (seconds * 1000)
28 |
29 |
30 | def convert_timestamp_ms_to_clock_time(timestamp_ms):
31 | """Converts time in milliseconds to clock time
32 | For instance, given 338000, it will return "00:05:38"
33 |
34 | Parameters
35 | ----------
36 | timestamp_ms : int
37 | Timestamp in milliseconds
38 |
39 | Returns
40 | -------
41 | clock_time : str
42 | The clock time in HH:mm:ss format
43 | """
44 | hours = int(timestamp_ms / 3600000)
45 | minute = int((timestamp_ms % 3600000) / 60000)
46 | seconds = float((timestamp_ms % 3600000 % 60000) / 1000)
47 |
48 | if int(seconds) == seconds:
49 | seconds = int(seconds)
50 |
51 | formatted_seconds = str(seconds)
52 | if seconds < 10:
53 | formatted_seconds = "0" + str(seconds)
54 |
55 | return str(hours).zfill(2) + ":" + str(minute).zfill(2) + ":" + formatted_seconds
56 |
--------------------------------------------------------------------------------
/src/video_segment_finder.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 | import cv2
3 |
4 |
5 | class PastFrameChangesTracker:
6 | """ A class that keeps track of changes from previous frames """
7 |
8 | def __init__(self):
9 | self.prev_frame_changes = [False, False, False, False, False]
10 |
11 | def are_previous_frames_stable(self):
12 | """Checks if all previous frames had no changes
13 |
14 | Returns
15 | -------
16 | is_stable : boolean
17 | True if all past frames had no changes; else False
18 | """
19 | return sum([1 if x else 0 for x in self.prev_frame_changes]) == 0
20 |
21 | def add_frame_change(self, has_changed):
22 | """Adds a change to the tracker
23 | If there are more than 5 items in the tracker, it will evict the oldest frame change
24 |
25 | Parameters
26 | ----------
27 | has_changed : boolean
28 | True if there was a change with the current frame vs the past frame; else False
29 |
30 | Returns
31 | -------
32 | is_stable : boolean
33 | True if all past frames had no changes; else False
34 | """
35 | self.prev_frame_changes.append(has_changed)
36 |
37 | if len(self.prev_frame_changes) > 5:
38 | self.prev_frame_changes.pop(0)
39 |
40 |
41 | class VideoSegmentFinder:
42 | """A class responsible for finding a list of best possible video segments
43 | A good video segment (a, t1, t2) is when image a is best explained when watching the video from time t1 to t2
44 |
45 | Attributes
46 | ----------
47 | threshold : int
48 | Is the min. difference between the color of two images on one pixel location for it to be distinct
49 | min_change : int
50 | Is the min. number of pixel changes between two adjacent video frames for the two to be considered distinct
51 | """
52 |
53 | def __init__(self, threshold=20, min_change=10000):
54 | self.threshold = threshold
55 | self.min_change = min_change
56 |
57 | def get_best_segment_frames(self, video_file):
58 | ''' Finds a list of best possible video segments
59 | It returns a map, where the key is the frame number, and the value is the frame data
60 |
61 | The frame data is of this format:
62 | {
63 | "timestamp": ,
64 | "frame": ,
65 | "next_frame": ,
66 | "mask": ,
67 | "num_pixels_changed": ,
68 | }
69 |
70 | The video segment can be obtained by two adjacent frame data, f1, f2 where:
71 | a = f2.frame
72 | t1 = f1.timestamp
73 | t2 = f2.timestamp
74 |
75 | Returns
76 | -------
77 | selected_frames : { a -> b }
78 | A map of frame number a to the frame data b
79 | '''
80 | selected_frames, _ = self.get_segment_frames_with_stats(
81 | video_file, save_stats_for_all_frames=False
82 | )
83 | return selected_frames
84 |
85 | def get_segment_frames_with_stats(self, video_file, save_stats_for_all_frames=True):
86 | ''' Returns a list of frames for the best possible video segments (refer to get_best_segment_frames())
87 |
88 | It also outputs statistics on all frames, where the statistic on frame i is:
89 | {
90 | "timestamp": the timestamp of frame i
91 | "num_pixels_changed": number of pixel changes from frame i - 1 to frame i
92 | }
93 |
94 | Returns
95 | -------
96 | selected_frames : { a -> b }
97 | A map of frame number to its frame data
98 | stats : { a -> c }
99 | A map of frame number to its statistic
100 | '''
101 |
102 | video_reader = cv2.VideoCapture(video_file)
103 |
104 | # Get the Default resolutions
105 | frame_width = int(video_reader.get(cv2.CAP_PROP_FRAME_WIDTH))
106 | frame_height = int(video_reader.get(cv2.CAP_PROP_FRAME_HEIGHT))
107 |
108 | # Get the FPS
109 | fps = int(video_reader.get(cv2.CAP_PROP_FPS))
110 |
111 | frame_num = 0
112 | frame_num_to_stats = {}
113 | selected_frames = {}
114 |
115 | prev_timestamp = 0
116 | prev_frame = 255 * np.ones(
117 | (frame_height, frame_width, 3), np.uint8
118 | ) # A blank screen
119 | prev_video_changes = PastFrameChangesTracker()
120 |
121 | while video_reader.isOpened():
122 | is_read, cur_frame = video_reader.read()
123 | timestamp = video_reader.get(cv2.CAP_PROP_POS_MSEC)
124 |
125 | # Is when the stream is ending
126 | if not is_read:
127 | break
128 |
129 | results = self.__compare_frames__(prev_frame, cur_frame)
130 |
131 | # Store the results
132 | if save_stats_for_all_frames:
133 | frame_num_to_stats[frame_num] = {
134 | "timestamp": timestamp,
135 | "num_pixels_changed": results["num_pixels_changed"],
136 | }
137 |
138 | has_changed = results["num_pixels_changed"] > self.min_change
139 | save_frame = False
140 |
141 | if prev_video_changes.are_previous_frames_stable() and has_changed:
142 | save_frame = True
143 |
144 | if save_frame:
145 | selected_frames[frame_num] = {
146 | "timestamp": prev_timestamp,
147 | "frame": prev_frame,
148 | "next_frame": cur_frame,
149 | "mask": results["mask"],
150 | "num_pixels_changed": results["num_pixels_changed"],
151 | }
152 |
153 | prev_video_changes.add_frame_change(has_changed)
154 |
155 | prev_frame = cur_frame
156 | prev_timestamp = timestamp
157 |
158 | frame_num += 1
159 |
160 | # Add the last frame of the video
161 | selected_frames[frame_num] = {
162 | "timestamp": prev_timestamp,
163 | "frame": prev_frame,
164 | "next_frame": 255
165 | * np.ones((frame_height, frame_width, 3), np.uint8), # A blank screen,
166 | "mask": prev_frame,
167 | "num_pixels_changed": 0,
168 | }
169 |
170 | # Rare case: if there are two selected frames s.t. they differ by 1 second, then there is a glitch
171 | # and we pick the frame that is the earliest
172 | selected_frame_nums = sorted(selected_frames.keys())
173 | i = 0
174 | while i < len(selected_frame_nums) - 1:
175 | cur_frame = selected_frames[selected_frame_nums[i]]
176 | next_frame = selected_frames[selected_frame_nums[i + 1]]
177 |
178 | if (next_frame["timestamp"] - cur_frame["timestamp"]) < 2000:
179 | del selected_frames[selected_frame_nums[i + 1]]
180 | i += 1
181 |
182 | i += 1
183 |
184 | # Edge case: delete the first selected frame since it is just a blank screen
185 | del selected_frames[selected_frame_nums[0]]
186 |
187 | video_reader.release()
188 | cv2.destroyAllWindows()
189 |
190 | return selected_frames, frame_num_to_stats
191 |
192 | def __compare_frames__(self, prev_frame, cur_frame):
193 | diff = cv2.absdiff(prev_frame, cur_frame)
194 | mask = cv2.cvtColor(diff, cv2.COLOR_BGR2GRAY)
195 | num_pixels_changed = np.sum(mask > self.threshold)
196 |
197 | return {"num_pixels_changed": num_pixels_changed, "mask": mask, "diff": diff}
198 |
199 |
200 | if __name__ == "__main__":
201 | splitter = VideoSegmentFinder()
202 | splitter.get_best_segment_frames("../tests/videos/input_2.mp4")
203 |
--------------------------------------------------------------------------------
/tests/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/__init__.py
--------------------------------------------------------------------------------
/tests/snapshots/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/__init__.py
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_1_with_srt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_1_with_srt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_1_with_webvtt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_1_with_webvtt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_1_without_subtitles.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_1_without_subtitles.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_2_with_srt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_2_with_srt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_2_with_webvtt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_2_with_webvtt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_2_without_subtitles.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_2_without_subtitles.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_7_with_srt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_7_with_srt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_7_without_subtitles.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_7_without_subtitles.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_8_with_srt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_8_with_srt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_8_with_webvtt_file.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_8_with_webvtt_file.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_main/video_8_without_subtitles.pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/snapshots/snap_test_main/video_8_without_subtitles.pdf
--------------------------------------------------------------------------------
/tests/snapshots/snap_test_subtitle_segment_finder.py:
--------------------------------------------------------------------------------
1 | # -*- coding: utf-8 -*-
2 | # snapshottest: v1 - https://goo.gl/zC4yUc
3 | from __future__ import unicode_literals
4 |
5 | from snapshottest import Snapshot
6 |
7 |
8 | snapshots = Snapshot()
9 |
10 | snapshots['SubtitleSplitterTests::test_get_pages_given_subtitle_1_should_return_correct_pages 1'] = [
11 | "So now that we understand what a secure PRG is, and we understand what semantic security means, we can actually argue that a stream cipher with a secure PRG is, in fact, a semantically secure. So that's our goal for this, segment. It's a fairly straightforward proof, and we'll see how it goes.",
12 | "So the theory we wanna prove is that, basically, given a generator G that happens to be a secured, psedo-random generator. In fact, the stream cipher that's derived from this generator is going to be semantically secure. Okay and I want to emphasize. That there was no hope of proving a theorem like this for perfect secrecy. For Shannons concept of perfect secrecy. Because we know that a stream cipher can not be perfectly secure because it has short keys. And perfect secrecy requires the keys to be as long as the message. So this is really kind of the first example the we see where we're able to prove that a cipher with short keys has security. The concept of security is semantic security. And this actually validates that, really, this is a very useful concept. And in fact, you know, we'll be using semantic security many, many times throughout the course. Okay, so how do we prove a theory like this? What we're actually gonna be doing, is we're gonna be proving the contrapositive. What we're gonna show is the following. So we're gonna prove this statement down here, but let me parse it for you. Suppose. You give me a semantic security adversary A. What we'll do is we'll build PRG adversary B to satisfy this inequality here. Now why is this inequality useful? Basically what do we know? We know that if B is an efficient adversary. Then we know that since G is a secure generator, we know that this advantage is negligible, right? A secure generator has a negligible advantage against any efficient statistical test. So the right hand side, basically, is gonna be negligible. But because the right hand side is negligible, we can deduce that the left hand side is negligible. And therefore, the adversary that you looked at actually has negligible advantage in attacking the stream cipher E. Okay. So this is how this, this will work. Basically all we have to do is given an adversary A we're going to build an adversary B. We know that B has negligible advantage against generator but that implies that A has negligible advantage against the stream cipher. So let's do that. So all we have to do again is given A, we have to build B.",
13 | "So let A be a semantic security adversary against the stream cipher. So let me remind you what that means. Basically, there's a challenger. The challenger starts off by choosing the key K. And then the adversary is gonna output two messages, two equal length messages. And he's gonna receive the encryption of M0 or M1 and outputs B1. Okay, that's what a semantic security adversary is going to do. So now we're going to start playing games with this adversary. And that's how we're going to prove our lemma. Alright, so the first thing we're going to do is we're going to make the challenger. Also choose a random R. Okay, a random string R. So, well you know the adversary doesn't really care what the challenger does internally. The challenger never uses R, so this doesn't affect the adversary's advantage at all. The adversary just doesn't care that the challenger also picks R. But now comes the trick. What we're going to do is we're going to, instead of encrypting using GK. We're going to encrypt using R. You can see basically what we're doing here. Essentially we're changing the challenger so now the challenge cipher text is encrypted using a truly random pad. As opposed to just pseudo random pad GK. Okay. Now, the property of the pseudo-random generator is that its output is indistinguishable from truly random. So, because the PRG is secure, the adversary can't tell that we made this change. The adversary can't tell that we switched from a pseudo-random string to a truly random string. Again, because the generator is secure. Well, but now look at the game that we ended up with. So the adversary's advantage couldn't have changed by much, because he can't tell the difference. But now look at the game that we ended up with. Now this game is truly a one time pad game. This a semantic security game against the one time pad. Because now the adversary is getting a one time pad encryption of M0 or M1 But in the one time pad we know that the adversaries advantage is zero, because you can't beat the one time pad. The one time pad is secure Unconditionally secure. And as a result, because of this. Essentially because the adversary couldn't have told the difference when we moved from pseudo random to random. But he couldn't win the random game. That also means that he couldn't win the sudo random game. And as a result, the stream cipher, the original stream cipher must be secure. So that's the intuition for how the proof is gonna go. But I wanna do it rigorously once. From now on, we're just gonna argue by playing games with our challenger. And, we won't be doing things as formal as I'm gonna do next. But I wanna do formally and precisely once, just so that you see how these proofs actually work. Okay, so I'm gonna have to introduce some notation. And I'll do the usual notation, basically. If the original semantics are here at the beginning, when we're actually using a pseudo-random pad, I'm gonna use W0 and W1 to denote the event that the adversary outputs one, when it gets the encryption of M0, or gets the encryption of M1, respectively. Okay? So W0 corresponds to outputting 1 when receiving the encryption of M0. And W1 corresponds to outputting 1 when receiving the encryption of M1. So that's the standard definition of semantic security. Now once we flip to the random pad. I'm gonna use R0 and R1 to denote the event that the adversary outputs 1 when receiving the one-type pad encryption of M0 or the one-time pad encryption of M1. So we have four events, W0, W1 from the original semmantics security game, and R0 and R1 from the semmantics security game once we switch over to the one-time pad.",
14 | "So now let's look at relations between these variables. So first of all, R0 and R1 are basically events from a semmantics security game against a one-time pad. So the difference between these probabilities is that, as we said, basically the advantage of algorithm A, of adversary A, against the one-time pad. Which we know is zero. Okay, so that's great. So that basically means that probability of, of R0 is equal to the probability of R1. So now, let's put these events on a line, on a line segment between zero and one. So here are the events. W0 and W1 are the events we're interested in. We wanna show that these two are close. Okay. And the way we're going to do it is basically by showing, oh and I should say, here is probability R0 and R1, it says they're both same, I just put them in the same place. What we're gonna do is we're gonna show that both W0 and W1 are actually close to the probability of RB and as a result they must be close to one another. Okay, so the way we do that is using a second claim, so now we're interested in the distance between probability of Wb and the probability of Rb. Okay so we'll prove the claim in a second. Let me just state the claim. The claim says that there exists in adversary B. Such that the difference of these two probabilities is basically the advantage of B against the generator G and this is for both b's. Okay? So given these two claims, like the theorem is done because basically what do we know. We know this distance is less than the advantage of B against G. That's from claim two and similarly, this distance actually is even equal to, I'm not gonna say less but is equal to the advantage. Of B against G, and as a result you can see that the distance between W0 and W1 is basically almost twice the advantage of B against G. That's basically the thing that we are trying to prove. Okay the only thing that remains is just proving this claim two and if you think about what claim two says, it basically captures the question of what happens in experiment zero what happens when we replace the pseudo random pad GK, by truly random pad R. Here in experiment zero say we're using the pseudo random pad and here in experiment zero we are using a Truly random pad and we are asking can the adversary tell the difference between these two and we wanna argue that he cannot because the generator is secure.",
15 | "Okay so here's what we are gonna do. So let's prove claim two. So we are gonna argue that in fact there is a PRG adversary B that has exactly the difference of the two probabilities as it's advantage. Okay and since the point is since this is negligible this is negligible. And that's basically what we wanted to prove. Okay, so let's look at the statistical test b. So, what, our statistical test b is gonna use adversary A in his belly, so we get to build statistical test b however we want. As we said, it's gonna use adversary A inside of it, for its operation, and it's a regular statistical test, so it takes an n-bit string as inputs, and it's supposed to output, you know, random or non-random, zero or one. Okay, so let's see. So it's, first thing it's gonna do, is it's gonna run adversary A, and adversary A is gonna output two messages, M0 and M1. And then, what adversary b's gonna do, is basically gonna respond. With M0 XOR or the string that it was given as inputs. Alright? That's the statistical lesson, then. Whenever A outputs, it's gonna output, its output. And now let's look at its advantage. So what can we say about the advantage of this statistical test against the generator? Well, so by definition, it's the probability that, if you choose a truly random string. So here are 01 to the N, so probability that R, that B outputs 1 minus the probability, is that when we choose a pseudo random string, B outputs 1, okay? Okay, but let's think about what this is. What can you tell me about the first expressions? What can you tell me about this expression over here? Well, by the definition that's exactly if you think about what's going on here, that's this is exactly the probability R0 right? Because this game that we are playing with the adversary here is basically he helped us M0 and M1 right here he helped add M0 and m1 and he got the encryption of M0 under truly one time pad. Okay, so this is basically a [inaudible]. Here let me write this a little better. That's the basic level probability of R0. Now, what can we say about the next expression, well what can we say about when B is given a pseudo random string Y as input. Well in that case, this is exactly experiment zero and true stream cipher game because now we're computing M XOR M0, XOR GK. This is exactly W0. Okay, that's exactly what we have to prove. So it's kind of a trivial proof. Okay, so that completes the proof of claim two.",
16 | "And again, just to make sure this is all clear, once we have claim two, we know that W0 must be close to W1, and that's the theorem. That's what we have to prove. Okay, so now we've established that a stream cypher is in fact symmantically secure, assuming that the PRG is secure.",
17 | ''
18 | ]
19 |
20 | snapshots['SubtitleSplitterTests::test_get_pages_given_subtitle_8_should_return_correct_pages 1'] = [
21 | '[',
22 | 'M',
23 | '',
24 | '',
25 | '',
26 | 'u',
27 | 'si',
28 | 'c]',
29 | 'well i welcome y',
30 | 'ou all to th',
31 | 'e lecture four of week',
32 | 'five in this week we will be studying the important building block of an automated system that is microprocessor',
33 | 'technology let us look at the outline of this lec',
34 | 'ture at start of the lecture we will see the definition of a micropro',
35 | 'cessor then we will see its architecture it has various elements how these el',
36 | 'ements operate then we will learn what is the difference between a microcontroller and a microprocessor the definition of microcontroller will be studied after that we will study the microcomputers which we are using for our regu',
37 | 'lar day-to-day activities at the end of the lecture we will study about the plcs the programmable logic controllers which are used in the automation industry its elements configuration and operation will be studied in detail fine',
38 | 'let us begin in this lecture we will be studying various programmable logic devices in a in our previous lectures we have seen the elements of measurement system such as the sensors and the signal conditioning devices after that we have also seen how we can convert the signals from one form to the another form now let us look at the fundamentals of programmable logic devices as the name suggests the programmable logic devices has two words these are programmable and logic the devices which are carrying out logical operations on the data are called as the logic devices but when we are able to program them when the users are able to program these logic devices then we are call them as the programmable logic devices programmable means we can teach we can instruct we can train these logic devices to carry out a certain set of instructions in the given sequence so what kind of operations these programmable logical logic devices are carrying out they are carrying out various control functions according to the instructions written in its memory so whatever the commands that we are giving they are the low level language commands we will see what is the meaning of the low level and the high level language in the next few',
39 | 'slides the first pld is the microprocessor microprocessor is a digital integrated circuit it is an electronic circuit and it carries out various digital functions which are necessary to process the information which are necessary to process the data given to the micropr',
40 | 'ocessor the microcomputer is utilizing the microprocessor as its cpu that is a central processing unit and it contains all functions of a computer so micro computer may have memory it can communicate with the outside world with input output devices so here you must know the difference between the processor and a computer processor is a simple circuit while computer is a system it has processor as the cpu central proc',
41 | 'essing unit third pld which is very widely used which is very important as far as the automated system is concerned is plc that is programmable logic controller programmable logic controller also incorporates the microprocessor as the central processing unit and it controls the operations of electromechanical devices but plc is working in very harsh condition so the construction of the plc must be very rugged very r',
42 | 'obust the general microprocessors which are used in automation industry are the embedded microprocessor some of the features of this microprocessor are first these kind of microprocessors are dedicated to a specific function control of a specific function for example we want to have an automatic control system to control the temperature of an electric furnace here the function is to control the temperature of the furnace if the temperature if the inside temperature of the furnace is above the set value the microprocessor should cut off the supply of electricity to that electric furnace so a simple electronic circuitry which is embedded inside the electric furnace is nothing but the embedded microprocessor so it is dedicated to carry out specific functions second feature of the embedded microprocessor is they are self starting so the processors are starting on their own as we switch on the system the processor will start on i',
43 | 'ts own why we are incorporating the microprocessor is in automation as we have seen the definition of automation that we want to reduce the human intervention we want to have no intervention so to carry out the operations in auto mode we are taking help of the electronic spec circuitry that is nothing',
44 | 'but the microprocessors these are completely self-contained the embedded microprocessor has their own memory has may have their own battery backup as well they may have their own energy sou',
45 | "rce as well the microprocessor which are used in automation which are embedded in in the products or systems has their own operating system so we can say an embedded microprocessor will be dedicated to a specific function it is self-starting it doesn't require any human intervention these are completely self-contained everything is there inside the system and they do have their own operating syst",
46 | 'em the microprocessor a multi-purpose programmable device it basically reads the binary instructions from a storage device that called memory the memory may be the temporary memory or it may be the permanent memory so whatever the signals which are getting in will be stored in a temporary memory the microprocessor process this information as per the need as per the instructions given in th',
47 | 'e program it process that information according to the instructions and it provides the results as the output so getting the information reading the information and processing the information based upon the instructions given are the functions of a microprocessor but how these functions are carried out to carry out these functions the microprocessor basically has three elements the processor memory element and input output devices to carry out these functions the microprocessor basically has hardware and this and the hardware is nothing but the electronic parts which are integrated together to carry out the intended operation the processor is an electrical circuitry memory is an electronic part and input output devices are are the electromechanical parts well these components are integrated together but their coordinated operation will be carried out by a set of instructions and that of set of instruction is nothing but the program and when we group together a variety of programs then we call that as a software so let us see what this processor does the processor recognizes the program instructions and it carry out it execute that program instructions as per the order given as per the sequence given the microprocessor has the in input output interfaces so the microprocessor is also carrying out the communication between the cpu that is the processor and the outside world through the input output interfaces so in general we call the interface as port in our day to day language as well we are using the word port quite often the third element which is there that is a memory the memory basically holds the program instructions and the data the applications of microprocessor can be classified basically in two categories so based upon the applications of the microprocessor we can say that the microprocessor can be utilized as a reprogrammable system device such as the microcomputer so whatever the computing devices that we do have we can reprogram these systems we can change its operating system we can install some programs we can modify uh these programs we can edit the programs so to carry out these operations in microcomputers we are using the microprocessor so when we are using the microprocessor utility for developing the reprogrammable system that we call the application of microprocessor to have the microcomputers in the second application we we want to have the dedicated functions that to be carried out by the microprocessor and these are nothing but the embedded systems which we have already talked about let us consider the typical automated system such as a conveyor belt automated guided vehicles automatic furnaces all the manufacturing industry equipment whether it may be processing equipment or it is the conveying equipment or it is the monitoring equipment so whenever we are saying we want to carry out the intended application the desired application in automatic mode there also we need the microprocessor so the application of microprocessor to carry out specific functions in automatic mode that is nothing but the embedded system these processors do operate in binary digits that is 0 and 1 these digits are called bits if the electrical voltage given to the machine is of low level then the bit would be 0 and if a high voltage is applied high electrical energy is applied then we are considering that as the beat 1.'
48 | ]
49 |
--------------------------------------------------------------------------------
/tests/subtitles/subtitles_1.srt:
--------------------------------------------------------------------------------
1 | 1
2 | 00:00:00,000 --> 00:00:04,134
3 | So now that we understand what a secure
4 | PRG is, and we understand what semantic
5 |
6 | 2
7 | 00:00:04,134 --> 00:00:08,425
8 | security means, we can actually argue that
9 | a stream cipher with a secure PRG is, in
10 |
11 | 3
12 | 00:00:08,425 --> 00:00:12,559
13 | fact, a semantically secure. So that's our
14 | goal for this, segment. It's a fairly
15 |
16 | 4
17 | 00:00:12,559 --> 00:00:16,746
18 | straightforward proof, and we'll see how
19 | it goes. So the theory we wanna prove is
20 |
21 | 5
22 | 00:00:16,746 --> 00:00:20,932
23 | that, basically, given a generator G that
24 | happens to be a secured, psedo-random
25 |
26 | 6
27 | 00:00:20,932 --> 00:00:24,805
28 | generator. In fact, the stream cipher
29 | that's derived from this generator is
30 |
31 | 7
32 | 00:00:24,805 --> 00:00:28,924
33 | going to be semantically secure. Okay and
34 | I want to emphasize. That there was no
35 |
36 | 8
37 | 00:00:28,924 --> 00:00:33,085
38 | hope of proving a theorem like this for
39 | perfect secrecy. For Shannons concept of
40 |
41 | 9
42 | 00:00:33,085 --> 00:00:37,193
43 | perfect secrecy. Because we know that a
44 | stream cipher can not be perfectly
45 |
46 | 10
47 | 00:00:37,193 --> 00:00:41,264
48 | secure because it has short keys. And
49 | perfect secrecy requires the keys to be as
50 |
51 | 11
52 | 00:00:41,264 --> 00:00:45,321
53 | long as the message. So this is really
54 | kind of the first example the we see where
55 |
56 | 12
57 | 00:00:45,321 --> 00:00:49,229
58 | we're able to prove that a cipher with
59 | short keys has security. The concept of
60 |
61 | 13
62 | 00:00:49,229 --> 00:00:53,236
63 | security is semantic security. And this
64 | actually validates that, really, this is a
65 |
66 | 14
67 | 00:00:53,236 --> 00:00:56,943
68 | very useful concept. And in fact, you
69 | know, we'll be using semantic security
70 |
71 | 15
72 | 00:00:56,943 --> 00:01:00,750
73 | many, many times throughout the course.
74 | Okay, so how do we prove a theory like
75 |
76 | 16
77 | 00:01:00,750 --> 00:01:04,257
78 | this? What we're actually gonna be doing,
79 | is we're gonna be proving the
80 |
81 | 17
82 | 00:01:04,257 --> 00:01:08,264
83 | contrapositive. What we're gonna show is
84 | the following. So we're gonna prove this
85 |
86 | 18
87 | 00:01:08,264 --> 00:01:12,815
88 | statement down here, but let me parse it
89 | for you. Suppose. You give me a semantic
90 |
91 | 19
92 | 00:01:12,815 --> 00:01:18,345
93 | security adversary A. What we'll do is
94 | we'll build PRG adversary B to satisfy
95 |
96 | 20
97 | 00:01:18,345 --> 00:01:23,686
98 | this inequality here. Now why is this
99 | inequality useful? Basically what do we
100 |
101 | 21
102 | 00:01:23,686 --> 00:01:28,878
103 | know? We know that if B is an efficient
104 | adversary. Then we know that since G is a
105 |
106 | 22
107 | 00:01:28,878 --> 00:01:33,053
108 | secure generator, we know that this
109 | advantage is negligible, right? A secure
110 |
111 | 23
112 | 00:01:33,053 --> 00:01:37,510
113 | generator has a negligible advantage
114 | against any efficient statistical test. So
115 |
116 | 24
117 | 00:01:37,510 --> 00:01:42,023
118 | the right hand side, basically, is gonna
119 | be negligible. But because the right hand
120 |
121 | 25
122 | 00:01:42,023 --> 00:01:46,023
123 | side is negligible, we can deduce that the
124 | left hand side is negligible.
125 |
126 | 26
127 | 00:01:46,023 --> 00:01:50,767
128 | And therefore, the adversary that you looked
129 | at actually has negligible advantage in
130 |
131 | 27
132 | 00:01:50,767 --> 00:01:54,538
133 | attacking the stream cipher E. Okay. So
134 | this is how this, this will work.
135 |
136 | 28
137 | 00:01:54,538 --> 00:01:58,486
138 | Basically all we have to do is given an
139 | adversary A we're going to build an
140 |
141 | 29
142 | 00:01:58,486 --> 00:02:02,591
143 | adversary B. We know that B has negligible
144 | advantage against generator but that
145 |
146 | 30
147 | 00:02:02,591 --> 00:02:06,036
148 | implies that A has negligible advantage
149 | against the stream cipher.
150 |
151 | 31
152 | 00:02:06,082 --> 00:02:10,994
153 | So let's do that. So all we have to do again
154 | is given A, we have to build B.
155 |
156 | 32
157 | 00:02:10,994 --> 00:02:15,183
158 | So let A be a semantic security adversary against
159 | the stream cipher. So let me remind you
160 |
161 | 33
162 | 00:02:15,183 --> 00:02:19,320
163 | what that means. Basically, there's a
164 | challenger. The challenger starts off by
165 |
166 | 34
167 | 00:02:19,320 --> 00:02:23,509
168 | choosing the key K. And then the adversary
169 | is gonna output two messages, two equal
170 |
171 | 35
172 | 00:02:23,509 --> 00:02:27,383
173 | length messages. And he's gonna receive
174 | the encryption of M0 or M1
175 |
176 | 36
177 | 00:02:27,383 --> 00:02:31,226
178 | and outputs B1. Okay, that's
179 | what a semantic security adversary is
180 |
181 | 37
182 | 00:02:31,226 --> 00:02:34,933
183 | going to do. So now we're going to start
184 | playing games with this adversary. And
185 |
186 | 38
187 | 00:02:34,933 --> 00:02:38,498
188 | that's how we're going to prove our lemma. Alright, so the first thing
189 |
190 | 39
191 | 00:02:38,498 --> 00:02:42,535
192 | we're going to do is we're going to make
193 | the challenger. Also choose a random R.
194 |
195 | 40
196 | 00:02:42,535 --> 00:02:47,500
197 | Okay, a random string R. So, well you know
198 | the adversary doesn't really care what the
199 |
200 | 41
201 | 00:02:47,500 --> 00:02:52,405
202 | challenger does internally. The challenger
203 | never uses R, so this doesn't affect the
204 |
205 | 42
206 | 00:02:52,405 --> 00:02:56,365
207 | adversary's advantage at all. The
208 | adversary just doesn't care that the
209 |
210 | 43
211 | 00:02:56,365 --> 00:03:00,706
212 | challenger also picks R. But now comes the
213 | trick. What we're going to do is we're
214 |
215 | 44
216 | 00:03:00,706 --> 00:03:05,042
217 | going to, instead of encrypting using GK.
218 | We're going to encrypt using R. You can
219 |
220 | 45
221 | 00:03:05,042 --> 00:03:09,993
222 | see basically what we're doing
223 | here. Essentially we're changing the
224 |
225 | 46
226 | 00:03:09,993 --> 00:03:14,219
227 | challenger so now the challenge
228 | cipher text is encrypted using a
229 |
230 | 47
231 | 00:03:14,219 --> 00:03:19,006
232 | truly random pad. As opposed to just pseudo
233 | random pad GK. Okay. Now, the property of
234 |
235 | 48
236 | 00:03:19,006 --> 00:03:23,639
237 | the pseudo-random generator is that its
238 | output is indistinguishable from truly
239 |
240 | 49
241 | 00:03:23,639 --> 00:03:28,273
242 | random. So, because the PRG is secure, the
243 | adversary can't tell that we made this
244 |
245 | 50
246 | 00:03:28,273 --> 00:03:33,082
247 | change. The adversary can't tell that we
248 | switched from a pseudo-random string to a
249 |
250 | 51
251 | 00:03:33,082 --> 00:03:37,422
252 | truly random string. Again, because the generator is secure. Well, but now look at
253 |
254 | 52
255 | 00:03:37,422 --> 00:03:41,762
256 | the game that we ended up with. So the
257 | adversary's advantage couldn't have
258 |
259 | 53
260 | 00:03:41,762 --> 00:03:46,630
261 | changed by much, because he can't tell the
262 | difference. But now look at the game that
263 |
264 | 54
265 | 00:03:46,630 --> 00:03:51,073
266 | we ended up with. Now this game is truly a
267 | one time pad game. This a semantic
268 |
269 | 55
270 | 00:03:51,073 --> 00:03:55,803
271 | security game against the one time pad.
272 | Because now the adversary is getting a one
273 |
274 | 56
275 | 00:03:55,803 --> 00:04:00,238
276 | time pad encryption of M0 or M1 But in the
277 | one time pad we know that the adversaries
278 |
279 | 57
280 | 00:04:00,238 --> 00:04:04,048
281 | advantage is zero, because you can't beat
282 | the one time pad. The one time pad is
283 |
284 | 58
285 | 00:04:04,048 --> 00:04:08,165
286 | secure Unconditionally secure. And as a
287 | result, because of this. Essentially
288 |
289 | 59
290 | 00:04:08,165 --> 00:04:12,674
291 | because the adversary couldn't have told
292 | the difference when
293 |
294 | 60
295 | 00:04:12,674 --> 00:04:17,013
296 | we moved from pseudo random to random. But he couldn't win the
297 | random game. That also means that he
298 |
299 | 61
300 | 00:04:17,013 --> 00:04:21,411
301 | couldn't win the sudo random game. And as
302 | a result, the stream cipher, the original
303 |
304 | 62
305 | 00:04:21,411 --> 00:04:25,634
306 | stream cipher must be secure. So that's
307 | the intuition for how the proof is gonna go.
308 |
309 | 63
310 | 00:04:25,634 --> 00:04:29,594
311 | But I wanna do it rigorously once.
312 | From now on, we're just gonna argue by
313 |
314 | 64
315 | 00:04:29,594 --> 00:04:33,975
316 | playing games with our challenger. And, we
317 | won't be doing things as formal as I'm
318 |
319 | 65
320 | 00:04:33,975 --> 00:04:38,304
321 | gonna do next. But I wanna do formally and
322 | precisely once, just so that you see how
323 |
324 | 66
325 | 00:04:38,304 --> 00:04:42,629
326 | these proofs actually work. Okay, so I'm
327 | gonna have to introduce some notation. And
328 |
329 | 67
330 | 00:04:42,629 --> 00:04:47,751
331 | I'll do the usual notation, basically. If
332 | the original semantics are here at the
333 |
334 | 68
335 | 00:04:47,751 --> 00:04:52,937
336 | beginning, when we're actually using a
337 | pseudo-random pad, I'm gonna use W0 and W1
338 |
339 | 69
340 | 00:04:52,937 --> 00:04:58,059
341 | to denote the event that the adversary
342 | outputs one, when it gets the encryption
343 |
344 | 70
345 | 00:04:58,059 --> 00:05:02,856
346 | of M0, or gets the encryption of M1,
347 | respectively. Okay? So W0 corresponds to
348 |
349 | 71
350 | 00:05:02,856 --> 00:05:07,719
351 | outputting 1 when receiving the
352 | encryption of M0. And W1 corresponds
353 |
354 | 72
355 | 00:05:07,719 --> 00:05:13,141
356 | to outputting 1 when receiving the encryption of M1. So that's the standard
357 |
358 | 73
359 | 00:05:13,141 --> 00:05:19,606
360 | definition of semantic security. Now once
361 | we flip to the random pad. I'm gonna use
362 |
363 | 74
364 | 00:05:19,606 --> 00:05:24,505
365 | R0 and R1 to denote the event that the
366 | adversary outputs 1 when receiving the
367 |
368 | 75
369 | 00:05:24,505 --> 00:05:29,125
370 | one-type pad encryption of M0 or the
371 | one-time pad encryption of M1. So we have
372 |
373 | 76
374 | 00:05:29,125 --> 00:05:33,567
375 | four events, W0, W1 from the original
376 | semmantics security game, and R0 and R1
377 |
378 | 77
379 | 00:05:33,567 --> 00:05:38,365
380 | from the semmantics security game once we
381 | switch over to the one-time pad. So now
382 |
383 | 78
384 | 00:05:38,365 --> 00:05:42,985
385 | let's look at relations between these
386 | variables. So first of all, R0 and R1 are
387 |
388 | 79
389 | 00:05:42,985 --> 00:05:47,427
390 | basically events from a semmantics
391 | security game against a one-time pad. So
392 |
393 | 80
394 | 00:05:47,427 --> 00:05:51,929
395 | the difference between these probabilities
396 | is that, as we said, basically the
397 |
398 | 81
399 | 00:05:51,929 --> 00:05:56,902
400 | advantage of algorithm A, of adversary A,
401 | against the one-time pad. Which we know is
402 |
403 | 82
404 | 00:05:56,902 --> 00:06:01,231
405 | zero. Okay, so that's great. So that
406 | basically means that probability of, of R0
407 |
408 | 83
409 | 00:06:01,231 --> 00:06:05,662
410 | is equal to the probability of R1. So now,
411 | let's put these events on a line, on a
412 |
413 | 84
414 | 00:06:05,662 --> 00:06:10,261
415 | line segment between zero and one. So here
416 | are the events. W0 and W1 are the events
417 |
418 | 85
419 | 00:06:10,261 --> 00:06:14,499
420 | we're interested in. We wanna show that
421 | these two are close. Okay. And the way
422 |
423 | 86
424 | 00:06:14,499 --> 00:06:18,490
425 | we're going to do it is basically by
426 | showing, oh and I should say, here is
427 |
428 | 87
429 | 00:06:18,490 --> 00:06:22,754
430 | probability R0 and R1, it says
431 | they're both same, I just put them in the
432 |
433 | 88
434 | 00:06:22,754 --> 00:06:27,237
435 | same place. What we're gonna do is we're
436 | gonna show that both W0 and W1 are
437 |
438 | 89
439 | 00:06:27,237 --> 00:06:31,720
440 | actually close to the probability of RB
441 | and as a result they must be close to one
442 |
443 | 90
444 | 00:06:31,720 --> 00:06:35,656
445 | another. Okay, so the way we do that is
446 | using a second claim, so now we're
447 |
448 | 91
449 | 00:06:35,656 --> 00:06:39,865
450 | interested in the distance between
451 | probability of Wb and the probability of
452 |
453 | 92
454 | 00:06:39,865 --> 00:06:44,730
455 | Rb. Okay so we'll prove the claim in a
456 | second. Let me just state the claim. The
457 |
458 | 93
459 | 00:06:44,730 --> 00:06:49,938
460 | claim says that there exists in adversary
461 | B. Such that the difference of these two
462 |
463 | 94
464 | 00:06:49,938 --> 00:06:55,081
465 | probabilities is basically the advantage
466 | of B against the generator G and this is
467 |
468 | 95
469 | 00:06:55,081 --> 00:06:59,833
470 | for both b's. Okay? So given these two
471 | claims, like the theorem is done because
472 |
473 | 96
474 | 00:06:59,833 --> 00:07:04,771
475 | basically what do we know. We know this
476 | distance is less than the advantage of B
477 |
478 | 97
479 | 00:07:04,771 --> 00:07:09,523
480 | against G. That's from claim two and
481 | similarly, this distance actually is even
482 |
483 | 98
484 | 00:07:09,523 --> 00:07:14,401
485 | equal to, I'm not gonna say
486 | less but is equal to the advantage. Of B
487 |
488 | 99
489 | 00:07:14,401 --> 00:07:19,455
490 | against G, and as a result you can see
491 | that the distance between W0 and W1
492 |
493 | 100
494 | 00:07:19,455 --> 00:07:24,446
495 | is basically almost twice the
496 | advantage of B against G. That's basically
497 |
498 | 101
499 | 00:07:24,446 --> 00:07:29,375
500 | the thing that we are trying to prove.
501 | Okay the only thing that remains is just
502 |
503 | 102
504 | 00:07:29,375 --> 00:07:34,304
505 | proving this claim two and if you think
506 | about what claim two says, it basically
507 |
508 | 103
509 | 00:07:34,304 --> 00:07:39,170
510 | captures the question of what happens in
511 | experiment zero what happens when we
512 |
513 | 104
514 | 00:07:39,170 --> 00:07:43,530
515 | replace the pseudo random pad GK, by
516 | truly random pad R. Here in
517 |
518 | 105
519 | 00:07:43,530 --> 00:07:48,910
520 | experiment zero say we're using the pseudo
521 | random pad and here in experiment zero we
522 |
523 | 106
524 | 00:07:48,910 --> 00:07:53,593
525 | are using a Truly random pad and we are
526 | asking can the adversary tell the
527 |
528 | 107
529 | 00:07:53,593 --> 00:07:58,972
530 | difference between these two and we wanna
531 | argue that he cannot because the generator
532 |
533 | 108
534 | 00:07:58,972 --> 00:08:03,845
535 | is secure. Okay so here's what we are
536 | gonna do. So let's prove claim two. So we
537 |
538 | 109
539 | 00:08:03,845 --> 00:08:08,728
540 | are gonna argue that in fact there is a
541 | PRG adversary B that has exactly the
542 |
543 | 110
544 | 00:08:08,728 --> 00:08:13,764
545 | difference of the two probabilities as
546 | it's advantage. Okay and since the point
547 |
548 | 111
549 | 00:08:13,764 --> 00:08:18,319
550 | is since this is negligible this is
551 | negligible. And that's basically what we
552 |
553 | 112
554 | 00:08:18,319 --> 00:08:22,323
555 | wanted to prove. Okay, so let's look at
556 | the statistical test b. So, what, our
557 |
558 | 113
559 | 00:08:22,323 --> 00:08:26,514
560 | statistical test b is gonna use adversary
561 | A in his belly, so we get to build
562 |
563 | 114
564 | 00:08:26,514 --> 00:08:31,091
565 | statistical test b however we want. As we
566 | said, it's gonna use adversary A inside of
567 |
568 | 115
569 | 00:08:31,091 --> 00:08:35,558
570 | it, for its operation, and it's a regular
571 | statistical test, so it takes an n-bit
572 |
573 | 116
574 | 00:08:35,558 --> 00:08:39,694
575 | string as inputs, and it's supposed to
576 | output, you know, random or non-random,
577 |
578 | 117
579 | 00:08:39,694 --> 00:08:43,995
580 | zero or one. Okay, so let's see. So it's,
581 | first thing it's gonna do, is it's gonna
582 |
583 | 118
584 | 00:08:43,995 --> 00:08:48,407
585 | run adversary A, and adversary A is gonna
586 | output two messages, M0 and M1. And then,
587 |
588 | 119
589 | 00:08:48,407 --> 00:08:54,053
590 | what adversary b's gonna do, is basically
591 | gonna respond. With M0 XOR or the string that
592 |
593 | 120
594 | 00:08:54,053 --> 00:08:59,942
595 | it was given as inputs. Alright? That's
596 | the statistical lesson, then. Whenever A
597 |
598 | 121
599 | 00:08:59,942 --> 00:09:05,418
600 | outputs, it's gonna output, its output.
601 | And now let's look at its advantage. So
602 |
603 | 122
604 | 00:09:05,418 --> 00:09:10,477
605 | what can we say about the advantage of
606 | this statistical test against the
607 |
608 | 123
609 | 00:09:10,477 --> 00:09:15,606
610 | generator? Well, so by definition, it's
611 | the probability that, if you choose a
612 |
613 | 124
614 | 00:09:15,606 --> 00:09:20,527
615 | truly random string. So here are 01 to the N, so probability
616 |
617 | 125
618 | 00:09:20,527 --> 00:09:25,587
619 | that R, that B outputs 1 minus the
620 | probability, is that when we choose a
621 |
622 | 126
623 | 00:09:25,587 --> 00:09:32,603
624 | pseudo random string, B outputs 1, okay?
625 | Okay, but let's think about what this is.
626 |
627 | 127
628 | 00:09:32,603 --> 00:09:37,398
629 | What can you tell me about the first
630 | expressions? What can you tell me about
631 |
632 | 128
633 | 00:09:37,398 --> 00:09:42,256
634 | this expression over here? Well, by the
635 | definition that's exactly if you think
636 |
637 | 129
638 | 00:09:42,256 --> 00:09:47,366
639 | about what's going on here, that's this is
640 | exactly the probability R0 right?
641 |
642 | 130
643 | 00:09:47,366 --> 00:09:52,729
644 | Because this game that we are playing with
645 | the adversary here is basically he helped
646 |
647 | 131
648 | 00:09:52,729 --> 00:09:58,028
649 | us M0 and M1 right here he helped add M0 and m1 and he got the encryption
650 |
651 | 132
652 | 00:09:58,028 --> 00:10:03,919
653 | of M0 under truly one time pad. Okay,
654 | so this is basically a [inaudible]. Here
655 |
656 | 133
657 | 00:10:03,919 --> 00:10:10,214
658 | let me write this a little better. That's
659 | the basic level probability of R0.
660 |
661 | 134
662 | 00:10:10,660 --> 00:10:15,467
663 | Now, what can we say about the next
664 | expression, well what can we say about
665 |
666 | 135
667 | 00:10:15,467 --> 00:10:19,100
668 | when B is given a pseudo random
669 | string Y as input.
670 |
671 | 136
672 | 00:10:19,100 --> 00:10:23,493
673 | Well in that case, this is exactly experiment zero and
674 | true stream cipher game
675 |
676 | 137
677 | 00:10:23,493 --> 00:10:29,999
678 | because now we're computing M XOR M0, XOR GK. This is
679 | exactly W0.
680 |
681 | 138
682 | 00:10:29,999 --> 00:10:33,015
683 | Okay, that's exactly what we have to prove. So it's kind of a trivial proof.
684 |
685 | 139
686 | 00:10:33,015 --> 00:10:39,563
687 | Okay, so that completes the proof of claim two. And again, just to
688 | make sure this is all clear, once we have
689 |
690 | 140
691 | 00:10:39,563 --> 00:10:43,799
692 | claim two, we know that W0 must be close
693 | to W1, and that's the theorem.
694 |
695 | 141
696 | 00:10:43,814 --> 00:10:50,383
697 | That's what we have to prove. Okay, so now we've
698 | established that a stream cypher is in
699 |
700 | 142
701 | 00:10:50,383 --> 00:10:53,880
702 | fact symmantically secure, assuming that
703 | the PRG is secure.
704 |
705 |
--------------------------------------------------------------------------------
/tests/subtitles/subtitles_1.vtt:
--------------------------------------------------------------------------------
1 | WEBVTT
2 |
3 | 1
4 | 00:00:00.000 --> 00:00:04.134
5 | So now that we understand what a secure
6 | PRG is, and we understand what semantic
7 |
8 | 2
9 | 00:00:04.134 --> 00:00:08.425
10 | security means, we can actually argue that
11 | a stream cipher with a secure PRG is, in
12 |
13 | 3
14 | 00:00:08.425 --> 00:00:12.559
15 | fact, a semantically secure. So that's our
16 | goal for this, segment. It's a fairly
17 |
18 | 4
19 | 00:00:12.559 --> 00:00:16.746
20 | straightforward proof, and we'll see how
21 | it goes. So the theory we wanna prove is
22 |
23 | 5
24 | 00:00:16.746 --> 00:00:20.932
25 | that, basically, given a generator G that
26 | happens to be a secured, psedo-random
27 |
28 | 6
29 | 00:00:20.932 --> 00:00:24.805
30 | generator. In fact, the stream cipher
31 | that's derived from this generator is
32 |
33 | 7
34 | 00:00:24.805 --> 00:00:28.924
35 | going to be semantically secure. Okay and
36 | I want to emphasize. That there was no
37 |
38 | 8
39 | 00:00:28.924 --> 00:00:33.085
40 | hope of proving a theorem like this for
41 | perfect secrecy. For Shannons concept of
42 |
43 | 9
44 | 00:00:33.085 --> 00:00:37.193
45 | perfect secrecy. Because we know that a
46 | stream cipher can not be perfectly
47 |
48 | 10
49 | 00:00:37.193 --> 00:00:41.264
50 | secure because it has short keys. And
51 | perfect secrecy requires the keys to be as
52 |
53 | 11
54 | 00:00:41.264 --> 00:00:45.321
55 | long as the message. So this is really
56 | kind of the first example the we see where
57 |
58 | 12
59 | 00:00:45.321 --> 00:00:49.229
60 | we're able to prove that a cipher with
61 | short keys has security. The concept of
62 |
63 | 13
64 | 00:00:49.229 --> 00:00:53.236
65 | security is semantic security. And this
66 | actually validates that, really, this is a
67 |
68 | 14
69 | 00:00:53.236 --> 00:00:56.943
70 | very useful concept. And in fact, you
71 | know, we'll be using semantic security
72 |
73 | 15
74 | 00:00:56.943 --> 00:01:00.750
75 | many, many times throughout the course.
76 | Okay, so how do we prove a theory like
77 |
78 | 16
79 | 00:01:00.750 --> 00:01:04.257
80 | this? What we're actually gonna be doing,
81 | is we're gonna be proving the
82 |
83 | 17
84 | 00:01:04.257 --> 00:01:08.264
85 | contrapositive. What we're gonna show is
86 | the following. So we're gonna prove this
87 |
88 | 18
89 | 00:01:08.264 --> 00:01:12.815
90 | statement down here, but let me parse it
91 | for you. Suppose. You give me a semantic
92 |
93 | 19
94 | 00:01:12.815 --> 00:01:18.345
95 | security adversary A. What we'll do is
96 | we'll build PRG adversary B to satisfy
97 |
98 | 20
99 | 00:01:18.345 --> 00:01:23.686
100 | this inequality here. Now why is this
101 | inequality useful? Basically what do we
102 |
103 | 21
104 | 00:01:23.686 --> 00:01:28.878
105 | know? We know that if B is an efficient
106 | adversary. Then we know that since G is a
107 |
108 | 22
109 | 00:01:28.878 --> 00:01:33.053
110 | secure generator, we know that this
111 | advantage is negligible, right? A secure
112 |
113 | 23
114 | 00:01:33.053 --> 00:01:37.510
115 | generator has a negligible advantage
116 | against any efficient statistical test. So
117 |
118 | 24
119 | 00:01:37.510 --> 00:01:42.023
120 | the right hand side, basically, is gonna
121 | be negligible. But because the right hand
122 |
123 | 25
124 | 00:01:42.023 --> 00:01:46.023
125 | side is negligible, we can deduce that the
126 | left hand side is negligible.
127 |
128 | 26
129 | 00:01:46.023 --> 00:01:50.767
130 | And therefore, the adversary that you looked
131 | at actually has negligible advantage in
132 |
133 | 27
134 | 00:01:50.767 --> 00:01:54.538
135 | attacking the stream cipher E. Okay. So
136 | this is how this, this will work.
137 |
138 | 28
139 | 00:01:54.538 --> 00:01:58.486
140 | Basically all we have to do is given an
141 | adversary A we're going to build an
142 |
143 | 29
144 | 00:01:58.486 --> 00:02:02.591
145 | adversary B. We know that B has negligible
146 | advantage against generator but that
147 |
148 | 30
149 | 00:02:02.591 --> 00:02:06.036
150 | implies that A has negligible advantage
151 | against the stream cipher.
152 |
153 | 31
154 | 00:02:06.082 --> 00:02:10.994
155 | So let's do that. So all we have to do again
156 | is given A, we have to build B.
157 |
158 | 32
159 | 00:02:10.994 --> 00:02:15.183
160 | So let A be a semantic security adversary against
161 | the stream cipher. So let me remind you
162 |
163 | 33
164 | 00:02:15.183 --> 00:02:19.320
165 | what that means. Basically, there's a
166 | challenger. The challenger starts off by
167 |
168 | 34
169 | 00:02:19.320 --> 00:02:23.509
170 | choosing the key K. And then the adversary
171 | is gonna output two messages, two equal
172 |
173 | 35
174 | 00:02:23.509 --> 00:02:27.383
175 | length messages. And he's gonna receive
176 | the encryption of M0 or M1
177 |
178 | 36
179 | 00:02:27.383 --> 00:02:31.226
180 | and outputs B1. Okay, that's
181 | what a semantic security adversary is
182 |
183 | 37
184 | 00:02:31.226 --> 00:02:34.933
185 | going to do. So now we're going to start
186 | playing games with this adversary. And
187 |
188 | 38
189 | 00:02:34.933 --> 00:02:38.498
190 | that's how we're going to prove our lemma. Alright, so the first thing
191 |
192 | 39
193 | 00:02:38.498 --> 00:02:42.535
194 | we're going to do is we're going to make
195 | the challenger. Also choose a random R.
196 |
197 | 40
198 | 00:02:42.535 --> 00:02:47.500
199 | Okay, a random string R. So, well you know
200 | the adversary doesn't really care what the
201 |
202 | 41
203 | 00:02:47.500 --> 00:02:52.405
204 | challenger does internally. The challenger
205 | never uses R, so this doesn't affect the
206 |
207 | 42
208 | 00:02:52.405 --> 00:02:56.365
209 | adversary's advantage at all. The
210 | adversary just doesn't care that the
211 |
212 | 43
213 | 00:02:56.365 --> 00:03:00.706
214 | challenger also picks R. But now comes the
215 | trick. What we're going to do is we're
216 |
217 | 44
218 | 00:03:00.706 --> 00:03:05.042
219 | going to, instead of encrypting using GK.
220 | We're going to encrypt using R. You can
221 |
222 | 45
223 | 00:03:05.042 --> 00:03:09.993
224 | see basically what we're doing
225 | here. Essentially we're changing the
226 |
227 | 46
228 | 00:03:09.993 --> 00:03:14.219
229 | challenger so now the challenge
230 | cipher text is encrypted using a
231 |
232 | 47
233 | 00:03:14.219 --> 00:03:19.006
234 | truly random pad. As opposed to just pseudo
235 | random pad GK. Okay. Now, the property of
236 |
237 | 48
238 | 00:03:19.006 --> 00:03:23.639
239 | the pseudo-random generator is that its
240 | output is indistinguishable from truly
241 |
242 | 49
243 | 00:03:23.639 --> 00:03:28.273
244 | random. So, because the PRG is secure, the
245 | adversary can't tell that we made this
246 |
247 | 50
248 | 00:03:28.273 --> 00:03:33.082
249 | change. The adversary can't tell that we
250 | switched from a pseudo-random string to a
251 |
252 | 51
253 | 00:03:33.082 --> 00:03:37.422
254 | truly random string. Again, because the generator is secure. Well, but now look at
255 |
256 | 52
257 | 00:03:37.422 --> 00:03:41.762
258 | the game that we ended up with. So the
259 | adversary's advantage couldn't have
260 |
261 | 53
262 | 00:03:41.762 --> 00:03:46.630
263 | changed by much, because he can't tell the
264 | difference. But now look at the game that
265 |
266 | 54
267 | 00:03:46.630 --> 00:03:51.073
268 | we ended up with. Now this game is truly a
269 | one time pad game. This a semantic
270 |
271 | 55
272 | 00:03:51.073 --> 00:03:55.803
273 | security game against the one time pad.
274 | Because now the adversary is getting a one
275 |
276 | 56
277 | 00:03:55.803 --> 00:04:00.238
278 | time pad encryption of M0 or M1 But in the
279 | one time pad we know that the adversaries
280 |
281 | 57
282 | 00:04:00.238 --> 00:04:04.048
283 | advantage is zero, because you can't beat
284 | the one time pad. The one time pad is
285 |
286 | 58
287 | 00:04:04.048 --> 00:04:08.165
288 | secure Unconditionally secure. And as a
289 | result, because of this. Essentially
290 |
291 | 59
292 | 00:04:08.165 --> 00:04:12.674
293 | because the adversary couldn't have told
294 | the difference when
295 |
296 | 60
297 | 00:04:12.674 --> 00:04:17.013
298 | we moved from pseudo random to random. But he couldn't win the
299 | random game. That also means that he
300 |
301 | 61
302 | 00:04:17.013 --> 00:04:21.411
303 | couldn't win the sudo random game. And as
304 | a result, the stream cipher, the original
305 |
306 | 62
307 | 00:04:21.411 --> 00:04:25.634
308 | stream cipher must be secure. So that's
309 | the intuition for how the proof is gonna go.
310 |
311 | 63
312 | 00:04:25.634 --> 00:04:29.594
313 | But I wanna do it rigorously once.
314 | From now on, we're just gonna argue by
315 |
316 | 64
317 | 00:04:29.594 --> 00:04:33.975
318 | playing games with our challenger. And, we
319 | won't be doing things as formal as I'm
320 |
321 | 65
322 | 00:04:33.975 --> 00:04:38.304
323 | gonna do next. But I wanna do formally and
324 | precisely once, just so that you see how
325 |
326 | 66
327 | 00:04:38.304 --> 00:04:42.629
328 | these proofs actually work. Okay, so I'm
329 | gonna have to introduce some notation. And
330 |
331 | 67
332 | 00:04:42.629 --> 00:04:47.751
333 | I'll do the usual notation, basically. If
334 | the original semantics are here at the
335 |
336 | 68
337 | 00:04:47.751 --> 00:04:52.937
338 | beginning, when we're actually using a
339 | pseudo-random pad, I'm gonna use W0 and W1
340 |
341 | 69
342 | 00:04:52.937 --> 00:04:58.059
343 | to denote the event that the adversary
344 | outputs one, when it gets the encryption
345 |
346 | 70
347 | 00:04:58.059 --> 00:05:02.856
348 | of M0, or gets the encryption of M1,
349 | respectively. Okay? So W0 corresponds to
350 |
351 | 71
352 | 00:05:02.856 --> 00:05:07.719
353 | outputting 1 when receiving the
354 | encryption of M0. And W1 corresponds
355 |
356 | 72
357 | 00:05:07.719 --> 00:05:13.141
358 | to outputting 1 when receiving the encryption of M1. So that's the standard
359 |
360 | 73
361 | 00:05:13.141 --> 00:05:19.606
362 | definition of semantic security. Now once
363 | we flip to the random pad. I'm gonna use
364 |
365 | 74
366 | 00:05:19.606 --> 00:05:24.505
367 | R0 and R1 to denote the event that the
368 | adversary outputs 1 when receiving the
369 |
370 | 75
371 | 00:05:24.505 --> 00:05:29.125
372 | one-type pad encryption of M0 or the
373 | one-time pad encryption of M1. So we have
374 |
375 | 76
376 | 00:05:29.125 --> 00:05:33.567
377 | four events, W0, W1 from the original
378 | semmantics security game, and R0 and R1
379 |
380 | 77
381 | 00:05:33.567 --> 00:05:38.365
382 | from the semmantics security game once we
383 | switch over to the one-time pad. So now
384 |
385 | 78
386 | 00:05:38.365 --> 00:05:42.985
387 | let's look at relations between these
388 | variables. So first of all, R0 and R1 are
389 |
390 | 79
391 | 00:05:42.985 --> 00:05:47.427
392 | basically events from a semmantics
393 | security game against a one-time pad. So
394 |
395 | 80
396 | 00:05:47.427 --> 00:05:51.929
397 | the difference between these probabilities
398 | is that, as we said, basically the
399 |
400 | 81
401 | 00:05:51.929 --> 00:05:56.902
402 | advantage of algorithm A, of adversary A,
403 | against the one-time pad. Which we know is
404 |
405 | 82
406 | 00:05:56.902 --> 00:06:01.231
407 | zero. Okay, so that's great. So that
408 | basically means that probability of, of R0
409 |
410 | 83
411 | 00:06:01.231 --> 00:06:05.662
412 | is equal to the probability of R1. So now,
413 | let's put these events on a line, on a
414 |
415 | 84
416 | 00:06:05.662 --> 00:06:10.261
417 | line segment between zero and one. So here
418 | are the events. W0 and W1 are the events
419 |
420 | 85
421 | 00:06:10.261 --> 00:06:14.499
422 | we're interested in. We wanna show that
423 | these two are close. Okay. And the way
424 |
425 | 86
426 | 00:06:14.499 --> 00:06:18.490
427 | we're going to do it is basically by
428 | showing, oh and I should say, here is
429 |
430 | 87
431 | 00:06:18.490 --> 00:06:22.754
432 | probability R0 and R1, it says
433 | they're both same, I just put them in the
434 |
435 | 88
436 | 00:06:22.754 --> 00:06:27.237
437 | same place. What we're gonna do is we're
438 | gonna show that both W0 and W1 are
439 |
440 | 89
441 | 00:06:27.237 --> 00:06:31.720
442 | actually close to the probability of RB
443 | and as a result they must be close to one
444 |
445 | 90
446 | 00:06:31.720 --> 00:06:35.656
447 | another. Okay, so the way we do that is
448 | using a second claim, so now we're
449 |
450 | 91
451 | 00:06:35.656 --> 00:06:39.865
452 | interested in the distance between
453 | probability of Wb and the probability of
454 |
455 | 92
456 | 00:06:39.865 --> 00:06:44.730
457 | Rb. Okay so we'll prove the claim in a
458 | second. Let me just state the claim. The
459 |
460 | 93
461 | 00:06:44.730 --> 00:06:49.938
462 | claim says that there exists in adversary
463 | B. Such that the difference of these two
464 |
465 | 94
466 | 00:06:49.938 --> 00:06:55.081
467 | probabilities is basically the advantage
468 | of B against the generator G and this is
469 |
470 | 95
471 | 00:06:55.081 --> 00:06:59.833
472 | for both b's. Okay? So given these two
473 | claims, like the theorem is done because
474 |
475 | 96
476 | 00:06:59.833 --> 00:07:04.771
477 | basically what do we know. We know this
478 | distance is less than the advantage of B
479 |
480 | 97
481 | 00:07:04.771 --> 00:07:09.523
482 | against G. That's from claim two and
483 | similarly, this distance actually is even
484 |
485 | 98
486 | 00:07:09.523 --> 00:07:14.401
487 | equal to, I'm not gonna say
488 | less but is equal to the advantage. Of B
489 |
490 | 99
491 | 00:07:14.401 --> 00:07:19.455
492 | against G, and as a result you can see
493 | that the distance between W0 and W1
494 |
495 | 100
496 | 00:07:19.455 --> 00:07:24.446
497 | is basically almost twice the
498 | advantage of B against G. That's basically
499 |
500 | 101
501 | 00:07:24.446 --> 00:07:29.375
502 | the thing that we are trying to prove.
503 | Okay the only thing that remains is just
504 |
505 | 102
506 | 00:07:29.375 --> 00:07:34.304
507 | proving this claim two and if you think
508 | about what claim two says, it basically
509 |
510 | 103
511 | 00:07:34.304 --> 00:07:39.170
512 | captures the question of what happens in
513 | experiment zero what happens when we
514 |
515 | 104
516 | 00:07:39.170 --> 00:07:43.530
517 | replace the pseudo random pad GK, by
518 | truly random pad R. Here in
519 |
520 | 105
521 | 00:07:43.530 --> 00:07:48.910
522 | experiment zero say we're using the pseudo
523 | random pad and here in experiment zero we
524 |
525 | 106
526 | 00:07:48.910 --> 00:07:53.593
527 | are using a Truly random pad and we are
528 | asking can the adversary tell the
529 |
530 | 107
531 | 00:07:53.593 --> 00:07:58.972
532 | difference between these two and we wanna
533 | argue that he cannot because the generator
534 |
535 | 108
536 | 00:07:58.972 --> 00:08:03.845
537 | is secure. Okay so here's what we are
538 | gonna do. So let's prove claim two. So we
539 |
540 | 109
541 | 00:08:03.845 --> 00:08:08.728
542 | are gonna argue that in fact there is a
543 | PRG adversary B that has exactly the
544 |
545 | 110
546 | 00:08:08.728 --> 00:08:13.764
547 | difference of the two probabilities as
548 | it's advantage. Okay and since the point
549 |
550 | 111
551 | 00:08:13.764 --> 00:08:18.319
552 | is since this is negligible this is
553 | negligible. And that's basically what we
554 |
555 | 112
556 | 00:08:18.319 --> 00:08:22.323
557 | wanted to prove. Okay, so let's look at
558 | the statistical test b. So, what, our
559 |
560 | 113
561 | 00:08:22.323 --> 00:08:26.514
562 | statistical test b is gonna use adversary
563 | A in his belly, so we get to build
564 |
565 | 114
566 | 00:08:26.514 --> 00:08:31.091
567 | statistical test b however we want. As we
568 | said, it's gonna use adversary A inside of
569 |
570 | 115
571 | 00:08:31.091 --> 00:08:35.558
572 | it, for its operation, and it's a regular
573 | statistical test, so it takes an n-bit
574 |
575 | 116
576 | 00:08:35.558 --> 00:08:39.694
577 | string as inputs, and it's supposed to
578 | output, you know, random or non-random,
579 |
580 | 117
581 | 00:08:39.694 --> 00:08:43.995
582 | zero or one. Okay, so let's see. So it's,
583 | first thing it's gonna do, is it's gonna
584 |
585 | 118
586 | 00:08:43.995 --> 00:08:48.407
587 | run adversary A, and adversary A is gonna
588 | output two messages, M0 and M1. And then,
589 |
590 | 119
591 | 00:08:48.407 --> 00:08:54.053
592 | what adversary b's gonna do, is basically
593 | gonna respond. With M0 XOR or the string that
594 |
595 | 120
596 | 00:08:54.053 --> 00:08:59.942
597 | it was given as inputs. Alright? That's
598 | the statistical lesson, then. Whenever A
599 |
600 | 121
601 | 00:08:59.942 --> 00:09:05.418
602 | outputs, it's gonna output, its output.
603 | And now let's look at its advantage. So
604 |
605 | 122
606 | 00:09:05.418 --> 00:09:10.477
607 | what can we say about the advantage of
608 | this statistical test against the
609 |
610 | 123
611 | 00:09:10.477 --> 00:09:15.606
612 | generator? Well, so by definition, it's
613 | the probability that, if you choose a
614 |
615 | 124
616 | 00:09:15.606 --> 00:09:20.527
617 | truly random string. So here are 01 to the N, so probability
618 |
619 | 125
620 | 00:09:20.527 --> 00:09:25.587
621 | that R, that B outputs 1 minus the
622 | probability, is that when we choose a
623 |
624 | 126
625 | 00:09:25.587 --> 00:09:32.603
626 | pseudo random string, B outputs 1, okay?
627 | Okay, but let's think about what this is.
628 |
629 | 127
630 | 00:09:32.603 --> 00:09:37.398
631 | What can you tell me about the first
632 | expressions? What can you tell me about
633 |
634 | 128
635 | 00:09:37.398 --> 00:09:42.256
636 | this expression over here? Well, by the
637 | definition that's exactly if you think
638 |
639 | 129
640 | 00:09:42.256 --> 00:09:47.366
641 | about what's going on here, that's this is
642 | exactly the probability R0 right?
643 |
644 | 130
645 | 00:09:47.366 --> 00:09:52.729
646 | Because this game that we are playing with
647 | the adversary here is basically he helped
648 |
649 | 131
650 | 00:09:52.729 --> 00:09:58.028
651 | us M0 and M1 right here he helped add M0 and m1 and he got the encryption
652 |
653 | 132
654 | 00:09:58.028 --> 00:10:03.919
655 | of M0 under truly one time pad. Okay,
656 | so this is basically a [inaudible]. Here
657 |
658 | 133
659 | 00:10:03.919 --> 00:10:10.214
660 | let me write this a little better. That's
661 | the basic level probability of R0.
662 |
663 | 134
664 | 00:10:10.660 --> 00:10:15.467
665 | Now, what can we say about the next
666 | expression, well what can we say about
667 |
668 | 135
669 | 00:10:15.467 --> 00:10:19.100
670 | when B is given a pseudo random
671 | string Y as input.
672 |
673 | 136
674 | 00:10:19.100 --> 00:10:23.493
675 | Well in that case, this is exactly experiment zero and
676 | true stream cipher game
677 |
678 | 137
679 | 00:10:23.493 --> 00:10:29.999
680 | because now we're computing M XOR M0, XOR GK. This is
681 | exactly W0.
682 |
683 | 138
684 | 00:10:29.999 --> 00:10:33.015
685 | Okay, that's exactly what we have to prove. So it's kind of a trivial proof.
686 |
687 | 139
688 | 00:10:33.015 --> 00:10:39.563
689 | Okay, so that completes the proof of claim two. And again, just to
690 | make sure this is all clear, once we have
691 |
692 | 140
693 | 00:10:39.563 --> 00:10:43.799
694 | claim two, we know that W0 must be close
695 | to W1, and that's the theorem.
696 |
697 | 141
698 | 00:10:43.814 --> 00:10:50.383
699 | That's what we have to prove. Okay, so now we've
700 | established that a stream cypher is in
701 |
702 | 142
703 | 00:10:50.383 --> 00:10:53.880
704 | fact symmantically secure, assuming that
705 | the PRG is secure.
--------------------------------------------------------------------------------
/tests/subtitles/subtitles_2.srt:
--------------------------------------------------------------------------------
1 | 1
2 | 00:00:00,000 --> 00:00:04,528
3 | The next thing we're going to look at is
4 | how to compute modular large integers. So
5 |
6 | 2
7 | 00:00:04,528 --> 00:00:09,023
8 | the first question is how do we represent
9 | large integers in a computer? So that's
10 |
11 | 3
12 | 00:00:09,023 --> 00:00:13,615
13 | actually fairly straightforward. So
14 | imagine we're on a 64 bit machine, what we
15 |
16 | 4
17 | 00:00:13,615 --> 00:00:18,361
18 | would do is we would break the number we
19 | want to represent, into 32 bit buckets And
20 |
21 | 5
22 | 00:00:18,361 --> 00:00:22,686
23 | then, we will basically have these n/32 bit buckets, and together they will
24 |
25 | 6
26 | 00:00:22,686 --> 00:00:26,906
27 | represent the number that we want to store
28 | on the computer. Now, I should mention
29 |
30 | 7
31 | 00:00:26,906 --> 00:00:30,705
32 | that I'm only giving 64 bit registers as
33 | an example. In fact, many modern
34 |
35 | 8
36 | 00:00:30,705 --> 00:00:34,977
37 | processors have 128 bit registers or more,
38 | and you can even do multiplications on
39 |
40 | 9
41 | 00:00:34,977 --> 00:00:38,987
42 | them. So normally you would actually use
43 | much larger blocks than just 32 bits. The
44 |
45 | 10
46 | 00:00:38,987 --> 00:00:42,943
47 | reason, by the way, you want to limit
48 | yourself to 32 bits is so that you can
49 |
50 | 11
51 | 00:00:42,943 --> 00:00:46,952
52 | multiply two blocks together, and the
53 | result will still be less than 64 bits,
54 |
55 | 12
56 | 00:00:46,952 --> 00:00:51,189
57 | less than the word size on the machine. So
58 | now let's look at particular arithmetic
59 |
60 | 13
61 | 00:00:51,189 --> 00:00:54,788
62 | operations and see how long each one
63 | takes. So addition and subtraction
64 |
65 | 14
66 | 00:00:54,788 --> 00:00:58,742
67 | basically what you would do is that
68 | addition would carry or subtraction would
69 |
70 | 15
71 | 00:00:58,742 --> 00:01:03,000
72 | borrow and those are basically linear time
73 | operations. In other words, if you want to
74 |
75 | 16
76 | 00:01:03,000 --> 00:01:06,954
77 | add or subtract two n bit integers the
78 | running time is basically
79 |
80 | 17
81 | 00:01:06,954 --> 00:01:12,626
82 | linear in n. Multiplication
83 | naively will take quadratic time. In fact,
84 |
85 | 18
86 | 00:01:12,626 --> 00:01:16,676
87 | this is what's called the high school
88 | algorithm. This is what you kind of
89 |
90 | 19
91 | 00:01:16,676 --> 00:01:21,114
92 | learned in school, where if you think
93 | about this for a minute you'll see that,
94 |
95 | 20
96 | 00:01:21,114 --> 00:01:25,662
97 | that algorithm basically is quadratic in
98 | the length of the numbers that are being
99 |
100 | 21
101 | 00:01:25,662 --> 00:01:30,156
102 | multiplied. So a big surprise in the 1960s
103 | was an algorithm due to Karatsuba that
104 |
105 | 22
106 | 00:01:30,156 --> 00:01:34,150
107 | actually achieves much better than
108 | quadratic time in fact, it achieved a
109 |
110 | 23
111 | 00:01:34,150 --> 00:01:38,567
112 | running time of n to the 1.585. And
113 | there's actually no point in me showing
114 |
115 | 24
116 | 00:01:38,567 --> 00:01:43,166
117 | you how the algorithm actually worked,
118 | I'll just mention the main idea What
119 |
120 | 25
121 | 00:01:43,166 --> 00:01:48,071
122 | Karatsuba realized, is that in fact when
123 | you want to multiply two numbers, you can
124 |
125 | 26
126 | 00:01:48,071 --> 00:01:52,976
127 | write them as, you can take the first
128 | number x, write it as 2 to the b times
129 |
130 | 27
131 | 00:01:52,976 --> 00:01:57,882
132 | x2 plus x1. Where x2 and x1 are roughly
133 | the size of the square root of x. Okay, so
134 |
135 | 28
136 | 00:01:57,882 --> 00:02:02,910
137 | we can kind of break the number x into the
138 | left part of x and the right part of x.
139 |
140 | 29
141 | 00:02:02,910 --> 00:02:07,654
142 | And basically, you're writing x as if it
143 | was written base 2 to the b. So it's got
144 |
145 | 30
146 | 00:02:07,654 --> 00:02:12,398
147 | two digits base 2 to the b. And you do
148 | the same thing with, y. You write y base
149 |
150 | 31
151 | 00:02:12,398 --> 00:02:16,911
152 | 2 to the b. Again, you would write it
153 | as, the sum of the left half plus the
154 |
155 | 32
156 | 00:02:16,911 --> 00:02:21,540
157 | right half, And then, normally, if you try
158 | to do this multiplication, when you open
159 |
160 | 33
161 | 00:02:21,540 --> 00:02:27,486
162 | up the parentheses. You see that, this
163 | would require 4 multiplications, right?
164 |
165 | 34
166 | 00:02:27,486 --> 00:02:33,365
167 | It would require x2 times y2, x2 times y1,
168 | x1 times y2, and x1 times y1. What
169 |
170 | 35
171 | 00:02:33,365 --> 00:02:39,879
172 | Karatsuba realized is there's a way to do
173 | this multiplication of x by y using only
174 |
175 | 36
176 | 00:02:39,879 --> 00:02:45,847
177 | three multiplications of x1 x2 y1 y2. So it's just a big multiplication of x times y
178 |
179 | 37
180 | 00:02:45,847 --> 00:02:50,214
181 | only it takes three little multiplications. You can now recursively
182 |
183 | 38
184 | 00:02:50,214 --> 00:02:55,087
185 | apply exactly the same procedure to
186 | multiplying x2 by y2, and x2 by y1, and so
187 |
188 | 39
189 | 00:02:55,087 --> 00:02:59,960
190 | on and so forth. And you would get this
191 | recursive algorithm. And if you do the
192 |
193 | 40
194 | 00:02:59,960 --> 00:03:05,087
195 | recursive analysis, you will see that
196 | basically, you get a running time of n to the 1.585.
197 |
198 | 41
199 | 00:03:05,087 --> 00:03:13,640
200 | This number is basically, the 1.585 is basically, log of 3 base 2.
201 |
202 | 42
203 | 00:03:13,640 --> 00:03:17,836
204 | Surprisingly, it turns out that Karatsuba
205 | isn't even the best multiplication
206 |
207 | 43
208 | 00:03:17,836 --> 00:03:23,912
209 | algorithm out there. It turns out that, in fact, you can do multiplication in about nlog(n) time.
210 |
211 | 44
212 | 00:03:23,912 --> 00:03:28,678
213 | So you can do multiplication in almost linear time. However, this is an extremely asymptotic results.
214 |
215 | 45
216 | 00:03:28,678 --> 00:03:31,477
217 | The big O here hides very big constants. And as a
218 |
219 | 46
220 | 00:03:31,477 --> 00:03:35,452
221 | result, this algorithm only becomes
222 | practical when the numbers are absolutely
223 |
224 | 47
225 | 00:03:35,452 --> 00:03:39,152
226 | enormous. And so this algorithm is
227 | actually not used very often. But
228 |
229 | 48
230 | 00:03:39,152 --> 00:03:43,183
231 | Karatsuba's algorithm is very practical.
232 | And in fact, most crypto-libraries
233 |
234 | 49
235 | 00:03:43,183 --> 00:03:47,885
236 | implement Karatsuba's algorithm for
237 | multiplication. However, for simplicity
238 |
239 | 50
240 | 00:03:47,885 --> 00:03:51,923
241 | here, I'm just gonna ignore Karatsuba's
242 | algorithm, and just for simplicity, I'm
243 |
244 | 51
245 | 00:03:51,923 --> 00:03:55,960
246 | gonna assume that multiplication runs in
247 | quadratic time. But in your mind, you
248 |
249 | 52
250 | 00:03:55,960 --> 00:03:59,893
251 | should always be thinking all multiplication really is a little bit faster than quadratic.
252 |
253 | 53
254 | 00:03:59,893 --> 00:04:04,794
255 | And then the next question after multiplication is what about
256 |
257 | 54
258 | 00:04:04,794 --> 00:04:10,297
259 | division with remainder and it turns out
260 | that's also a quadratic time algorithm.
261 |
262 | 55
263 | 00:04:10,297 --> 00:04:15,420
264 | So the main operation that remains, and one
265 | that we've used many times so far, and
266 |
267 | 56
268 | 00:04:15,420 --> 00:04:20,346
269 | I've never, actually never, ever told you
270 | how to actually compute it, is this
271 |
272 | 57
273 | 00:04:20,346 --> 00:04:26,339
274 | question of exponentiation. So let's solve
275 | this exponentiation problem a bit more
276 |
277 | 58
278 | 00:04:26,339 --> 00:04:31,558
279 | abstractly. So imagine we have a finite
280 | cyclic group G. All this means is that
281 |
282 | 59
283 | 00:04:31,558 --> 00:04:37,115
284 | this group is generated from the powers of
285 | some generator little g. So for example
286 |
287 | 60
288 | 00:04:37,115 --> 00:04:42,673
289 | think of this group as simply ZP, and think of little g as some generator of
290 |
291 | 61
292 | 00:04:42,673 --> 00:04:48,886
293 | big G. The reason I'm sitting in this way,
294 | is I'm, I want you to start getting used
295 |
296 | 62
297 | 00:04:48,886 --> 00:04:54,023
298 | to this abstraction where we deal with a
299 | generic group G and ZP really is just
300 |
301 | 63
302 | 00:04:54,023 --> 00:04:58,915
303 | one example of such a group. But, in fact,
304 | there are many other examples of finite
305 |
306 | 64
307 | 00:04:58,915 --> 00:05:03,379
308 | cyclic groups. And again I want to
309 | emphasis basically that group G, all it
310 |
311 | 65
312 | 00:05:03,379 --> 00:05:08,087
313 | is, it's simply this powers of this
314 | generator up to the order of the group.
315 |
316 | 66
317 | 00:05:08,087 --> 00:05:15,153
318 | I'll write it as G to the Q. So our goal
319 | now is given this element g, and some
320 |
321 | 67
322 | 00:05:15,153 --> 00:05:20,797
323 | exponent x, our goal is to compute the
324 | value of g to the x. Now normally what you
325 |
326 | 68
327 | 00:05:20,797 --> 00:05:24,810
328 | would say is, you would think well, you
329 | know, if x is equal to 3 then I'm
330 |
331 | 69
332 | 00:05:24,810 --> 00:05:28,898
333 | gonna compute you know, g cubed. Well,
334 | there's really nothing to do. All I do is
335 |
336 | 70
337 | 00:05:28,898 --> 00:05:32,795
338 | I just do g times g times g and I get g
339 | cubed, which is what I wanted. So that's
340 |
341 | 71
342 | 00:05:32,795 --> 00:05:36,790
343 | actually pretty easy. But in fact, that's
344 | not the case that we're interested in. In
345 |
346 | 72
347 | 00:05:36,790 --> 00:05:40,638
348 | our case, our exponents are gonna be
349 | enormous. And so if you try, you know,
350 |
351 | 73
352 | 00:05:40,638 --> 00:05:45,644
353 | think of like a 500-bit number and so if
354 | you try to compute g to the power of a
355 |
356 | 74
357 | 00:05:45,644 --> 00:05:50,710
358 | 500-bit number simply by multiplying g by
359 | g by g by g this is gonna take quite a
360 |
361 | 75
362 | 00:05:50,710 --> 00:05:55,716
363 | while. In fact it will take exponential
364 | time which is not something that we want
365 |
366 | 76
367 | 00:05:55,897 --> 00:06:00,722
368 | to do. So the question is whether even
369 | though x is enormous, can we still compute
370 |
371 | 77
372 | 00:06:00,722 --> 00:06:05,667
373 | g to the x relatively fast and the answer
374 | is yes and the algorithm that does that
375 |
376 | 78
377 | 00:06:05,667 --> 00:06:10,822
378 | is called a repeated squaring algorithm.
379 | And so let me show you how repeated
380 |
381 | 79
382 | 00:06:10,822 --> 00:06:15,593
383 | squaring works. So let's take as an
384 | example, 53. Naively you would have to do
385 |
386 | 80
387 | 00:06:15,593 --> 00:06:20,295
388 | 53 multiplications of g by g by g by g
389 | until you get to g by the 53 but I want to
390 |
391 | 81
392 | 00:06:20,295 --> 00:06:25,275
393 | show you how you can do it very quickly.
394 | So what we'll do is we'll write 53 in
395 |
396 | 82
397 | 00:06:25,275 --> 00:06:30,497
398 | binary. So here this is the binary
399 | representation of 53. And all that means
400 |
401 | 83
402 | 00:06:30,497 --> 00:06:36,282
403 | is, you notice this one corresponds to 32,
404 | this one corresponds to 16, this one
405 |
406 | 84
407 | 00:06:36,282 --> 00:06:41,292
408 | corresponds to 4, and this one
409 | corresponds to 1. So really 53 is 32
410 |
411 | 85
412 | 00:06:41,292 --> 00:06:47,038
413 | plus 16 plus 4 plus 1. But what
414 | that means is that g to the power of 53 is
415 |
416 | 86
417 | 00:06:47,038 --> 00:06:51,801
418 | g to the power of 32+16+4+1. And we can
419 | break that up, using again, the rules of
420 |
421 | 87
422 | 00:06:51,801 --> 00:06:57,235
423 | exponentiation. We can break that up as g
424 | to the 32 times g to the 16 times g to the
425 |
426 | 88
427 | 00:06:57,235 --> 00:07:02,938
428 | 4 times g to the 1, Now that should start
429 | giving you an idea for how to compute g to
430 |
431 | 89
432 | 00:07:02,938 --> 00:07:07,141
433 | the 53 very quickly. What we'll do is,
434 | simply, we'll take g and we'll start
435 |
436 | 90
437 | 00:07:07,141 --> 00:07:11,459
438 | squaring it. So what square wants, g wants
439 | to get g squared. We square it again to
440 |
441 | 91
442 | 00:07:11,459 --> 00:07:15,778
443 | get g to the 4, turn g to the 8.
444 | Turn g to the 16, g to the 32. So
445 |
446 | 92
447 | 00:07:15,778 --> 00:07:20,607
448 | we've computed all these squares of g. And
449 | now, what we're gonna do is we're simply
450 |
451 | 93
452 | 00:07:20,607 --> 00:07:25,719
453 | gonna multiply the appropriate powers to
454 | give us the g to the 53. So this is g to
455 |
456 | 94
457 | 00:07:25,719 --> 00:07:30,390
458 | the one times g to the 4 times g to the 16 times g to the 32, is basically
459 |
460 | 95
461 | 00:07:30,390 --> 00:07:35,376
462 | gonna give us the value that we want,
463 | which is g to the 53. So here you see that
464 |
465 | 96
466 | 00:07:35,376 --> 00:07:40,173
467 | all we had to do was just compute, let's
468 | see, we had to do one, two, three, four,
469 |
470 | 97
471 | 00:07:40,173 --> 00:07:49,343
472 | five squaring, plus four more multiplications
473 | so with 9 multiplications we computed g
474 |
475 | 98
476 | 00:07:49,343 --> 00:07:53,726
477 | to the 53. Okay so that's pretty
478 | interesting. And it turns out this is a
479 |
480 | 99
481 | 00:07:53,726 --> 00:07:58,253
482 | general phenomena that allows us to raise
483 | g to very, very high powers and do it very
484 |
485 | 100
486 | 00:07:58,253 --> 00:08:02,509
487 | quickly. So let me show you the algorithm,
488 | as I said this is called the repeated
489 |
490 | 101
491 | 00:08:02,509 --> 00:08:06,497
492 | squaring algorithm. So the input to the
493 | algorithm is the element g and the
494 |
495 | 102
496 | 00:08:06,497 --> 00:08:10,858
497 | exponent x. And the output is g to the x.
498 | So what we're gonna do is we're gonna
499 |
500 | 103
501 | 00:08:10,858 --> 00:08:15,415
502 | write x in binary notation. So let's say
503 | that x has n bits. And this is the actual
504 |
505 | 104
506 | 00:08:15,415 --> 00:08:19,521
507 | bit representation of x as a binary
508 | number. And then what we'll do is the
509 |
510 | 105
511 | 00:08:19,521 --> 00:08:24,246
512 | following. We'll have these two registers.
513 | y is gonna be a register that's constantly
514 |
515 | 106
516 | 00:08:24,246 --> 00:08:28,127
517 | squared. And then z is gonna be an
518 | accumulator that multiplies in the
519 |
520 | 107
521 | 00:08:28,127 --> 00:08:32,683
522 | appropriate powers of g as needed. So all
523 | we do is the following we loop through the
524 |
525 | 108
526 | 00:08:32,683 --> 00:08:36,526
527 | bits of x starting from the least
528 | significant bits, And then we do the
529 |
530 | 109
531 | 00:08:36,526 --> 00:08:41,414
532 | following: in every iteration we're simply
533 | gonna square y. Okay, so y just keeps on
534 |
535 | 110
536 | 00:08:41,414 --> 00:08:45,940
537 | squaring at every iteration. And then
538 | whenever the corresponding bit of the
539 |
540 | 111
541 | 00:08:45,940 --> 00:08:50,554
542 | exponent x happens to be one, we simply
543 | accumulate the current value of y into
544 |
545 | 112
546 | 00:08:50,554 --> 00:08:55,173
547 | this accumulator z and then at the end, we
548 | simply output z. That's it. That's the
549 |
550 | 113
551 | 00:08:55,173 --> 00:08:59,558
552 | whole algorithm, and that's the repeated
553 | squaring algorithm. So, let's see an
554 |
555 | 114
556 | 00:08:59,558 --> 00:09:04,060
557 | example with G to the 53. So,
558 | you can see the two columns. y is one
559 |
560 | 115
561 | 00:09:04,060 --> 00:09:08,387
562 | column, as it evolves through the
563 | iterations, and z is another column, again
564 |
565 | 116
566 | 00:09:08,387 --> 00:09:13,064
567 | as it evolves through the iterations. So,
568 | y is not very interesting. Basically, all
569 |
570 | 117
571 | 00:09:13,064 --> 00:09:17,449
572 | that happens to y is that at every
573 | iteration, it simply gets squared. And so
574 |
575 | 118
576 | 00:09:17,449 --> 00:09:22,299
577 | it just walks through the powers of two
578 | and the exponents and that's it. z is the
579 |
580 | 119
581 | 00:09:22,299 --> 00:09:26,915
582 | more interesting register where what it
583 | does is it accumulates the appropriate
584 |
585 | 120
586 | 00:09:26,915 --> 00:09:31,882
587 | powers of g whenever the corresponding bit
588 | to the exponent is one. So for example the
589 |
590 | 121
591 | 00:09:31,882 --> 00:09:36,031
592 | first bit of the exponent is one,
593 | therefore, the, at the end of the first
594 |
595 | 122
596 | 00:09:36,031 --> 00:09:41,219
597 | iteration the value of z is simply equal to
598 | g. The second bit of the exponent is zero
599 |
600 | 123
601 | 00:09:41,219 --> 00:09:46,473
602 | so the value of z doesn't change after the
603 | second iteration. And at the end of the
604 |
605 | 124
606 | 00:09:46,473 --> 00:09:51,856
607 | third iteration well the third bit of the
608 | exponent is one so we accumulate g to the
609 |
610 | 125
611 | 00:09:51,856 --> 00:09:56,662
612 | fourth into z. The next bit of the
613 | exponent is zero so z doesn't change. The
614 |
615 | 126
616 | 00:09:56,662 --> 00:10:02,109
617 | next bit of the exponent is one and so now
618 | we're supposed to accumulate the previous
619 |
620 | 127
621 | 00:10:02,109 --> 00:10:07,491
622 | value of y into the accumulator z so let
623 | me ask you so what's gonna be the value of z?
624 |
625 | 128
626 | 00:10:10,868 --> 00:10:14,245
627 | Well, we simply accumulate g to the
628 | 16 into z and so we simply compute the sum
629 |
630 | 129
631 | 00:10:14,245 --> 00:10:19,594
632 | of 16 and 5 we get g to the 21.
633 | Finally, the last bit is also set to one
634 |
635 | 130
636 | 00:10:19,594 --> 00:10:24,943
637 | so we accumulate it into z, we do 32 plus 21 and we get the finally output g to the 53.
638 |
639 | 131
640 | 00:10:24,943 --> 00:10:30,022
641 | Okay, so this gives you an idea of how
642 | this repeated squaring algorithm works.
643 |
644 | 132
645 | 00:10:30,022 --> 00:10:35,777
646 | It's is quite an interesting algorithm and
647 | it allows us to compute enormous powers of
648 |
649 | 133
650 | 00:10:35,777 --> 00:10:41,064
651 | g very, very, very quickly. So the number
652 | of iterations here, essentially, would be
653 |
654 | 134
655 | 00:10:41,064 --> 00:10:46,456
656 | log base 2 of x. Okay. You notice the
657 | number of iterations simply depends on the
658 |
659 | 135
660 | 00:10:46,456 --> 00:10:51,954
661 | number of digits of x, which is basically
662 | the log base 2 of x. So even if x is a
663 |
664 | 136
665 | 00:10:51,954 --> 00:10:56,519
666 | 500 bit number in 500 multiplication,
667 | well, 500 iterations, really 1,000
668 |
669 | 137
670 | 00:10:56,519 --> 00:11:01,736
671 | multiplications because we have to square
672 | and we have to accumulate. So in 1,000
673 |
674 | 138
675 | 00:11:01,736 --> 00:11:06,627
676 | multiplications we'll be able to raise g
677 | to the power of a 500 bit exponent.
678 |
679 | 139
680 | 00:11:06,627 --> 00:11:12,760
681 | Okay so now we can summarize kind of the running times so suppose we
682 |
683 | 140
684 | 00:11:12,760 --> 00:11:17,680
685 | have an N bit modulus capital N as we
686 | said addition and subtraction in ZN takes
687 |
688 | 141
689 | 00:11:17,680 --> 00:11:22,157
690 | linear time. Multiplication of just, you
691 | know, as I said, Karatsuba's actually makes this
692 |
693 | 142
694 | 00:11:22,157 --> 00:11:26,897
695 | more efficient, but for simplicity we'll
696 | just say that it takes quadratic time. And
697 |
698 | 143
699 | 00:11:26,897 --> 00:11:31,579
700 | then exponentiation, as I said, basically
701 | takes log of x iterations, and at each
702 |
703 | 144
704 | 00:11:31,579 --> 00:11:35,509
705 | iteration we basically do two
706 | multiplications. So it's O(log (x))
707 |
708 | 145
709 | 00:11:35,509 --> 00:11:40,307
710 | times the time to multiply. And let's say
711 | that the time to multiply is quadratic. So
712 |
713 | 146
714 | 00:11:40,307 --> 00:11:44,758
715 | the running time would be, really, N
716 | squared log x. And since x is always gonna
717 |
718 | 147
719 | 00:11:44,758 --> 00:11:49,168
720 | be less than N, by Fermat's theorem
721 | there's no point in raising g to a power
722 |
723 | 148
724 | 00:11:49,168 --> 00:11:53,958
725 | that's larger than the modulus. So x is
726 | gonna be less than N. Let's suppose that x
727 |
728 | 149
729 | 00:11:53,958 --> 00:11:58,570
730 | is also an N-bit integer, then, really
731 | exponentiation is a cubic-time algorithm.
732 |
733 | 150
734 | 00:11:58,570 --> 00:12:02,970
735 | Okay so that's what I wanted you to
736 | remember, that exponentiation is actually
737 |
738 | 151
739 | 00:12:02,970 --> 00:12:07,163
740 | a relatively slow. These days, it actually
741 | takes a few microseconds on a modern
742 |
743 | 152
744 | 00:12:07,163 --> 00:12:11,259
745 | computer. But still, microseconds for a,
746 | for a, say a four gigahertz processor is
747 |
748 | 153
749 | 00:12:11,259 --> 00:12:15,092
750 | quite a bit of work. And so just keep in
751 | mind that all the exponentiation
752 |
753 | 154
754 | 00:12:15,092 --> 00:12:19,556
755 | operations we talked about. For example,
756 | for determining if a number is a quadratic
757 |
758 | 155
759 | 00:12:19,556 --> 00:12:23,600
760 | residue or not, All those, all those
761 | exponentiations basically take cubic time.
762 |
763 | 156
764 | 00:12:24,780 --> 00:12:29,677
765 | Okay, so that completes our discussion of
766 | arithmetic algorithms, and then in the
767 |
768 | 157
769 | 00:12:29,677 --> 00:12:34,078
770 | next segment we'll start talking about
771 | hard problems, modulo, primes and composites.
772 |
773 |
--------------------------------------------------------------------------------
/tests/subtitles/subtitles_2.vtt:
--------------------------------------------------------------------------------
1 | WEBVTT
2 |
3 | 1
4 | 00:00:00.000 --> 00:00:04.528
5 | The next thing we're going to look at is
6 | how to compute modular large integers. So
7 |
8 | 2
9 | 00:00:04.528 --> 00:00:09.023
10 | the first question is how do we represent
11 | large integers in a computer? So that's
12 |
13 | 3
14 | 00:00:09.023 --> 00:00:13.615
15 | actually fairly straightforward. So
16 | imagine we're on a 64 bit machine, what we
17 |
18 | 4
19 | 00:00:13.615 --> 00:00:18.361
20 | would do is we would break the number we
21 | want to represent, into 32 bit buckets And
22 |
23 | 5
24 | 00:00:18.361 --> 00:00:22.686
25 | then, we will basically have these n/32 bit buckets, and together they will
26 |
27 | 6
28 | 00:00:22.686 --> 00:00:26.906
29 | represent the number that we want to store
30 | on the computer. Now, I should mention
31 |
32 | 7
33 | 00:00:26.906 --> 00:00:30.705
34 | that I'm only giving 64 bit registers as
35 | an example. In fact, many modern
36 |
37 | 8
38 | 00:00:30.705 --> 00:00:34.977
39 | processors have 128 bit registers or more,
40 | and you can even do multiplications on
41 |
42 | 9
43 | 00:00:34.977 --> 00:00:38.987
44 | them. So normally you would actually use
45 | much larger blocks than just 32 bits. The
46 |
47 | 10
48 | 00:00:38.987 --> 00:00:42.943
49 | reason, by the way, you want to limit
50 | yourself to 32 bits is so that you can
51 |
52 | 11
53 | 00:00:42.943 --> 00:00:46.952
54 | multiply two blocks together, and the
55 | result will still be less than 64 bits,
56 |
57 | 12
58 | 00:00:46.952 --> 00:00:51.189
59 | less than the word size on the machine. So
60 | now let's look at particular arithmetic
61 |
62 | 13
63 | 00:00:51.189 --> 00:00:54.788
64 | operations and see how long each one
65 | takes. So addition and subtraction
66 |
67 | 14
68 | 00:00:54.788 --> 00:00:58.742
69 | basically what you would do is that
70 | addition would carry or subtraction would
71 |
72 | 15
73 | 00:00:58.742 --> 00:01:03.000
74 | borrow and those are basically linear time
75 | operations. In other words, if you want to
76 |
77 | 16
78 | 00:01:03.000 --> 00:01:06.954
79 | add or subtract two n bit integers the
80 | running time is basically
81 |
82 | 17
83 | 00:01:06.954 --> 00:01:12.626
84 | linear in n. Multiplication
85 | naively will take quadratic time. In fact,
86 |
87 | 18
88 | 00:01:12.626 --> 00:01:16.676
89 | this is what's called the high school
90 | algorithm. This is what you kind of
91 |
92 | 19
93 | 00:01:16.676 --> 00:01:21.114
94 | learned in school, where if you think
95 | about this for a minute you'll see that,
96 |
97 | 20
98 | 00:01:21.114 --> 00:01:25.662
99 | that algorithm basically is quadratic in
100 | the length of the numbers that are being
101 |
102 | 21
103 | 00:01:25.662 --> 00:01:30.156
104 | multiplied. So a big surprise in the 1960s
105 | was an algorithm due to Karatsuba that
106 |
107 | 22
108 | 00:01:30.156 --> 00:01:34.150
109 | actually achieves much better than
110 | quadratic time in fact, it achieved a
111 |
112 | 23
113 | 00:01:34.150 --> 00:01:38.567
114 | running time of n to the 1.585. And
115 | there's actually no point in me showing
116 |
117 | 24
118 | 00:01:38.567 --> 00:01:43.166
119 | you how the algorithm actually worked,
120 | I'll just mention the main idea What
121 |
122 | 25
123 | 00:01:43.166 --> 00:01:48.071
124 | Karatsuba realized, is that in fact when
125 | you want to multiply two numbers, you can
126 |
127 | 26
128 | 00:01:48.071 --> 00:01:52.976
129 | write them as, you can take the first
130 | number x, write it as 2 to the b times
131 |
132 | 27
133 | 00:01:52.976 --> 00:01:57.882
134 | x2 plus x1. Where x2 and x1 are roughly
135 | the size of the square root of x. Okay, so
136 |
137 | 28
138 | 00:01:57.882 --> 00:02:02.910
139 | we can kind of break the number x into the
140 | left part of x and the right part of x.
141 |
142 | 29
143 | 00:02:02.910 --> 00:02:07.654
144 | And basically, you're writing x as if it
145 | was written base 2 to the b. So it's got
146 |
147 | 30
148 | 00:02:07.654 --> 00:02:12.398
149 | two digits base 2 to the b. And you do
150 | the same thing with, y. You write y base
151 |
152 | 31
153 | 00:02:12.398 --> 00:02:16.911
154 | 2 to the b. Again, you would write it
155 | as, the sum of the left half plus the
156 |
157 | 32
158 | 00:02:16.911 --> 00:02:21.540
159 | right half, And then, normally, if you try
160 | to do this multiplication, when you open
161 |
162 | 33
163 | 00:02:21.540 --> 00:02:27.486
164 | up the parentheses. You see that, this
165 | would require 4 multiplications, right?
166 |
167 | 34
168 | 00:02:27.486 --> 00:02:33.365
169 | It would require x2 times y2, x2 times y1,
170 | x1 times y2, and x1 times y1. What
171 |
172 | 35
173 | 00:02:33.365 --> 00:02:39.879
174 | Karatsuba realized is there's a way to do
175 | this multiplication of x by y using only
176 |
177 | 36
178 | 00:02:39.879 --> 00:02:45.847
179 | three multiplications of x1 x2 y1 y2. So it's just a big multiplication of x times y
180 |
181 | 37
182 | 00:02:45.847 --> 00:02:50.214
183 | only it takes three little multiplications. You can now recursively
184 |
185 | 38
186 | 00:02:50.214 --> 00:02:55.087
187 | apply exactly the same procedure to
188 | multiplying x2 by y2, and x2 by y1, and so
189 |
190 | 39
191 | 00:02:55.087 --> 00:02:59.960
192 | on and so forth. And you would get this
193 | recursive algorithm. And if you do the
194 |
195 | 40
196 | 00:02:59.960 --> 00:03:05.087
197 | recursive analysis, you will see that
198 | basically, you get a running time of n to the 1.585.
199 |
200 | 41
201 | 00:03:05.087 --> 00:03:13.640
202 | This number is basically, the 1.585 is basically, log of 3 base 2.
203 |
204 | 42
205 | 00:03:13.640 --> 00:03:17.836
206 | Surprisingly, it turns out that Karatsuba
207 | isn't even the best multiplication
208 |
209 | 43
210 | 00:03:17.836 --> 00:03:23.912
211 | algorithm out there. It turns out that, in fact, you can do multiplication in about nlog(n) time.
212 |
213 | 44
214 | 00:03:23.912 --> 00:03:28.678
215 | So you can do multiplication in almost linear time. However, this is an extremely asymptotic results.
216 |
217 | 45
218 | 00:03:28.678 --> 00:03:31.477
219 | The big O here hides very big constants. And as a
220 |
221 | 46
222 | 00:03:31.477 --> 00:03:35.452
223 | result, this algorithm only becomes
224 | practical when the numbers are absolutely
225 |
226 | 47
227 | 00:03:35.452 --> 00:03:39.152
228 | enormous. And so this algorithm is
229 | actually not used very often. But
230 |
231 | 48
232 | 00:03:39.152 --> 00:03:43.183
233 | Karatsuba's algorithm is very practical.
234 | And in fact, most crypto-libraries
235 |
236 | 49
237 | 00:03:43.183 --> 00:03:47.885
238 | implement Karatsuba's algorithm for
239 | multiplication. However, for simplicity
240 |
241 | 50
242 | 00:03:47.885 --> 00:03:51.923
243 | here, I'm just gonna ignore Karatsuba's
244 | algorithm, and just for simplicity, I'm
245 |
246 | 51
247 | 00:03:51.923 --> 00:03:55.960
248 | gonna assume that multiplication runs in
249 | quadratic time. But in your mind, you
250 |
251 | 52
252 | 00:03:55.960 --> 00:03:59.893
253 | should always be thinking all multiplication really is a little bit faster than quadratic.
254 |
255 | 53
256 | 00:03:59.893 --> 00:04:04.794
257 | And then the next question after multiplication is what about
258 |
259 | 54
260 | 00:04:04.794 --> 00:04:10.297
261 | division with remainder and it turns out
262 | that's also a quadratic time algorithm.
263 |
264 | 55
265 | 00:04:10.297 --> 00:04:15.420
266 | So the main operation that remains, and one
267 | that we've used many times so far, and
268 |
269 | 56
270 | 00:04:15.420 --> 00:04:20.346
271 | I've never, actually never, ever told you
272 | how to actually compute it, is this
273 |
274 | 57
275 | 00:04:20.346 --> 00:04:26.339
276 | question of exponentiation. So let's solve
277 | this exponentiation problem a bit more
278 |
279 | 58
280 | 00:04:26.339 --> 00:04:31.558
281 | abstractly. So imagine we have a finite
282 | cyclic group G. All this means is that
283 |
284 | 59
285 | 00:04:31.558 --> 00:04:37.115
286 | this group is generated from the powers of
287 | some generator little g. So for example
288 |
289 | 60
290 | 00:04:37.115 --> 00:04:42.673
291 | think of this group as simply ZP, and think of little g as some generator of
292 |
293 | 61
294 | 00:04:42.673 --> 00:04:48.886
295 | big G. The reason I'm sitting in this way,
296 | is I'm, I want you to start getting used
297 |
298 | 62
299 | 00:04:48.886 --> 00:04:54.023
300 | to this abstraction where we deal with a
301 | generic group G and ZP really is just
302 |
303 | 63
304 | 00:04:54.023 --> 00:04:58.915
305 | one example of such a group. But, in fact,
306 | there are many other examples of finite
307 |
308 | 64
309 | 00:04:58.915 --> 00:05:03.379
310 | cyclic groups. And again I want to
311 | emphasis basically that group G, all it
312 |
313 | 65
314 | 00:05:03.379 --> 00:05:08.087
315 | is, it's simply this powers of this
316 | generator up to the order of the group.
317 |
318 | 66
319 | 00:05:08.087 --> 00:05:15.153
320 | I'll write it as G to the Q. So our goal
321 | now is given this element g, and some
322 |
323 | 67
324 | 00:05:15.153 --> 00:05:20.797
325 | exponent x, our goal is to compute the
326 | value of g to the x. Now normally what you
327 |
328 | 68
329 | 00:05:20.797 --> 00:05:24.810
330 | would say is, you would think well, you
331 | know, if x is equal to 3 then I'm
332 |
333 | 69
334 | 00:05:24.810 --> 00:05:28.898
335 | gonna compute you know, g cubed. Well,
336 | there's really nothing to do. All I do is
337 |
338 | 70
339 | 00:05:28.898 --> 00:05:32.795
340 | I just do g times g times g and I get g
341 | cubed, which is what I wanted. So that's
342 |
343 | 71
344 | 00:05:32.795 --> 00:05:36.790
345 | actually pretty easy. But in fact, that's
346 | not the case that we're interested in. In
347 |
348 | 72
349 | 00:05:36.790 --> 00:05:40.638
350 | our case, our exponents are gonna be
351 | enormous. And so if you try, you know,
352 |
353 | 73
354 | 00:05:40.638 --> 00:05:45.644
355 | think of like a 500-bit number and so if
356 | you try to compute g to the power of a
357 |
358 | 74
359 | 00:05:45.644 --> 00:05:50.710
360 | 500-bit number simply by multiplying g by
361 | g by g by g this is gonna take quite a
362 |
363 | 75
364 | 00:05:50.710 --> 00:05:55.716
365 | while. In fact it will take exponential
366 | time which is not something that we want
367 |
368 | 76
369 | 00:05:55.897 --> 00:06:00.722
370 | to do. So the question is whether even
371 | though x is enormous, can we still compute
372 |
373 | 77
374 | 00:06:00.722 --> 00:06:05.667
375 | g to the x relatively fast and the answer
376 | is yes and the algorithm that does that
377 |
378 | 78
379 | 00:06:05.667 --> 00:06:10.822
380 | is called a repeated squaring algorithm.
381 | And so let me show you how repeated
382 |
383 | 79
384 | 00:06:10.822 --> 00:06:15.593
385 | squaring works. So let's take as an
386 | example, 53. Naively you would have to do
387 |
388 | 80
389 | 00:06:15.593 --> 00:06:20.295
390 | 53 multiplications of g by g by g by g
391 | until you get to g by the 53 but I want to
392 |
393 | 81
394 | 00:06:20.295 --> 00:06:25.275
395 | show you how you can do it very quickly.
396 | So what we'll do is we'll write 53 in
397 |
398 | 82
399 | 00:06:25.275 --> 00:06:30.497
400 | binary. So here this is the binary
401 | representation of 53. And all that means
402 |
403 | 83
404 | 00:06:30.497 --> 00:06:36.282
405 | is, you notice this one corresponds to 32,
406 | this one corresponds to 16, this one
407 |
408 | 84
409 | 00:06:36.282 --> 00:06:41.292
410 | corresponds to 4, and this one
411 | corresponds to 1. So really 53 is 32
412 |
413 | 85
414 | 00:06:41.292 --> 00:06:47.038
415 | plus 16 plus 4 plus 1. But what
416 | that means is that g to the power of 53 is
417 |
418 | 86
419 | 00:06:47.038 --> 00:06:51.801
420 | g to the power of 32+16+4+1. And we can
421 | break that up, using again, the rules of
422 |
423 | 87
424 | 00:06:51.801 --> 00:06:57.235
425 | exponentiation. We can break that up as g
426 | to the 32 times g to the 16 times g to the
427 |
428 | 88
429 | 00:06:57.235 --> 00:07:02.938
430 | 4 times g to the 1, Now that should start
431 | giving you an idea for how to compute g to
432 |
433 | 89
434 | 00:07:02.938 --> 00:07:07.141
435 | the 53 very quickly. What we'll do is,
436 | simply, we'll take g and we'll start
437 |
438 | 90
439 | 00:07:07.141 --> 00:07:11.459
440 | squaring it. So what square wants, g wants
441 | to get g squared. We square it again to
442 |
443 | 91
444 | 00:07:11.459 --> 00:07:15.778
445 | get g to the 4, turn g to the 8.
446 | Turn g to the 16, g to the 32. So
447 |
448 | 92
449 | 00:07:15.778 --> 00:07:20.607
450 | we've computed all these squares of g. And
451 | now, what we're gonna do is we're simply
452 |
453 | 93
454 | 00:07:20.607 --> 00:07:25.719
455 | gonna multiply the appropriate powers to
456 | give us the g to the 53. So this is g to
457 |
458 | 94
459 | 00:07:25.719 --> 00:07:30.390
460 | the one times g to the 4 times g to the 16 times g to the 32, is basically
461 |
462 | 95
463 | 00:07:30.390 --> 00:07:35.376
464 | gonna give us the value that we want,
465 | which is g to the 53. So here you see that
466 |
467 | 96
468 | 00:07:35.376 --> 00:07:40.173
469 | all we had to do was just compute, let's
470 | see, we had to do one, two, three, four,
471 |
472 | 97
473 | 00:07:40.173 --> 00:07:49.343
474 | five squaring, plus four more multiplications
475 | so with 9 multiplications we computed g
476 |
477 | 98
478 | 00:07:49.343 --> 00:07:53.726
479 | to the 53. Okay so that's pretty
480 | interesting. And it turns out this is a
481 |
482 | 99
483 | 00:07:53.726 --> 00:07:58.253
484 | general phenomena that allows us to raise
485 | g to very, very high powers and do it very
486 |
487 | 100
488 | 00:07:58.253 --> 00:08:02.509
489 | quickly. So let me show you the algorithm,
490 | as I said this is called the repeated
491 |
492 | 101
493 | 00:08:02.509 --> 00:08:06.497
494 | squaring algorithm. So the input to the
495 | algorithm is the element g and the
496 |
497 | 102
498 | 00:08:06.497 --> 00:08:10.858
499 | exponent x. And the output is g to the x.
500 | So what we're gonna do is we're gonna
501 |
502 | 103
503 | 00:08:10.858 --> 00:08:15.415
504 | write x in binary notation. So let's say
505 | that x has n bits. And this is the actual
506 |
507 | 104
508 | 00:08:15.415 --> 00:08:19.521
509 | bit representation of x as a binary
510 | number. And then what we'll do is the
511 |
512 | 105
513 | 00:08:19.521 --> 00:08:24.246
514 | following. We'll have these two registers.
515 | y is gonna be a register that's constantly
516 |
517 | 106
518 | 00:08:24.246 --> 00:08:28.127
519 | squared. And then z is gonna be an
520 | accumulator that multiplies in the
521 |
522 | 107
523 | 00:08:28.127 --> 00:08:32.683
524 | appropriate powers of g as needed. So all
525 | we do is the following we loop through the
526 |
527 | 108
528 | 00:08:32.683 --> 00:08:36.526
529 | bits of x starting from the least
530 | significant bits, And then we do the
531 |
532 | 109
533 | 00:08:36.526 --> 00:08:41.414
534 | following: in every iteration we're simply
535 | gonna square y. Okay, so y just keeps on
536 |
537 | 110
538 | 00:08:41.414 --> 00:08:45.940
539 | squaring at every iteration. And then
540 | whenever the corresponding bit of the
541 |
542 | 111
543 | 00:08:45.940 --> 00:08:50.554
544 | exponent x happens to be one, we simply
545 | accumulate the current value of y into
546 |
547 | 112
548 | 00:08:50.554 --> 00:08:55.173
549 | this accumulator z and then at the end, we
550 | simply output z. That's it. That's the
551 |
552 | 113
553 | 00:08:55.173 --> 00:08:59.558
554 | whole algorithm, and that's the repeated
555 | squaring algorithm. So, let's see an
556 |
557 | 114
558 | 00:08:59.558 --> 00:09:04.060
559 | example with G to the 53. So,
560 | you can see the two columns. y is one
561 |
562 | 115
563 | 00:09:04.060 --> 00:09:08.387
564 | column, as it evolves through the
565 | iterations, and z is another column, again
566 |
567 | 116
568 | 00:09:08.387 --> 00:09:13.064
569 | as it evolves through the iterations. So,
570 | y is not very interesting. Basically, all
571 |
572 | 117
573 | 00:09:13.064 --> 00:09:17.449
574 | that happens to y is that at every
575 | iteration, it simply gets squared. And so
576 |
577 | 118
578 | 00:09:17.449 --> 00:09:22.299
579 | it just walks through the powers of two
580 | and the exponents and that's it. z is the
581 |
582 | 119
583 | 00:09:22.299 --> 00:09:26.915
584 | more interesting register where what it
585 | does is it accumulates the appropriate
586 |
587 | 120
588 | 00:09:26.915 --> 00:09:31.882
589 | powers of g whenever the corresponding bit
590 | to the exponent is one. So for example the
591 |
592 | 121
593 | 00:09:31.882 --> 00:09:36.031
594 | first bit of the exponent is one,
595 | therefore, the, at the end of the first
596 |
597 | 122
598 | 00:09:36.031 --> 00:09:41.219
599 | iteration the value of z is simply equal to
600 | g. The second bit of the exponent is zero
601 |
602 | 123
603 | 00:09:41.219 --> 00:09:46.473
604 | so the value of z doesn't change after the
605 | second iteration. And at the end of the
606 |
607 | 124
608 | 00:09:46.473 --> 00:09:51.856
609 | third iteration well the third bit of the
610 | exponent is one so we accumulate g to the
611 |
612 | 125
613 | 00:09:51.856 --> 00:09:56.662
614 | fourth into z. The next bit of the
615 | exponent is zero so z doesn't change. The
616 |
617 | 126
618 | 00:09:56.662 --> 00:10:02.109
619 | next bit of the exponent is one and so now
620 | we're supposed to accumulate the previous
621 |
622 | 127
623 | 00:10:02.109 --> 00:10:07.491
624 | value of y into the accumulator z so let
625 | me ask you so what's gonna be the value of z?
626 |
627 | 128
628 | 00:10:10.868 --> 00:10:14.245
629 | Well, we simply accumulate g to the
630 | 16 into z and so we simply compute the sum
631 |
632 | 129
633 | 00:10:14.245 --> 00:10:19.594
634 | of 16 and 5 we get g to the 21.
635 | Finally, the last bit is also set to one
636 |
637 | 130
638 | 00:10:19.594 --> 00:10:24.943
639 | so we accumulate it into z, we do 32 plus 21 and we get the finally output g to the 53.
640 |
641 | 131
642 | 00:10:24.943 --> 00:10:30.022
643 | Okay, so this gives you an idea of how
644 | this repeated squaring algorithm works.
645 |
646 | 132
647 | 00:10:30.022 --> 00:10:35.777
648 | It's is quite an interesting algorithm and
649 | it allows us to compute enormous powers of
650 |
651 | 133
652 | 00:10:35.777 --> 00:10:41.064
653 | g very, very, very quickly. So the number
654 | of iterations here, essentially, would be
655 |
656 | 134
657 | 00:10:41.064 --> 00:10:46.456
658 | log base 2 of x. Okay. You notice the
659 | number of iterations simply depends on the
660 |
661 | 135
662 | 00:10:46.456 --> 00:10:51.954
663 | number of digits of x, which is basically
664 | the log base 2 of x. So even if x is a
665 |
666 | 136
667 | 00:10:51.954 --> 00:10:56.519
668 | 500 bit number in 500 multiplication,
669 | well, 500 iterations, really 1,000
670 |
671 | 137
672 | 00:10:56.519 --> 00:11:01.736
673 | multiplications because we have to square
674 | and we have to accumulate. So in 1,000
675 |
676 | 138
677 | 00:11:01.736 --> 00:11:06.627
678 | multiplications we'll be able to raise g
679 | to the power of a 500 bit exponent.
680 |
681 | 139
682 | 00:11:06.627 --> 00:11:12.760
683 | Okay so now we can summarize kind of the running times so suppose we
684 |
685 | 140
686 | 00:11:12.760 --> 00:11:17.680
687 | have an N bit modulus capital N as we
688 | said addition and subtraction in ZN takes
689 |
690 | 141
691 | 00:11:17.680 --> 00:11:22.157
692 | linear time. Multiplication of just, you
693 | know, as I said, Karatsuba's actually makes this
694 |
695 | 142
696 | 00:11:22.157 --> 00:11:26.897
697 | more efficient, but for simplicity we'll
698 | just say that it takes quadratic time. And
699 |
700 | 143
701 | 00:11:26.897 --> 00:11:31.579
702 | then exponentiation, as I said, basically
703 | takes log of x iterations, and at each
704 |
705 | 144
706 | 00:11:31.579 --> 00:11:35.509
707 | iteration we basically do two
708 | multiplications. So it's O(log (x))
709 |
710 | 145
711 | 00:11:35.509 --> 00:11:40.307
712 | times the time to multiply. And let's say
713 | that the time to multiply is quadratic. So
714 |
715 | 146
716 | 00:11:40.307 --> 00:11:44.758
717 | the running time would be, really, N
718 | squared log x. And since x is always gonna
719 |
720 | 147
721 | 00:11:44.758 --> 00:11:49.168
722 | be less than N, by Fermat's theorem
723 | there's no point in raising g to a power
724 |
725 | 148
726 | 00:11:49.168 --> 00:11:53.958
727 | that's larger than the modulus. So x is
728 | gonna be less than N. Let's suppose that x
729 |
730 | 149
731 | 00:11:53.958 --> 00:11:58.570
732 | is also an N-bit integer, then, really
733 | exponentiation is a cubic-time algorithm.
734 |
735 | 150
736 | 00:11:58.570 --> 00:12:02.970
737 | Okay so that's what I wanted you to
738 | remember, that exponentiation is actually
739 |
740 | 151
741 | 00:12:02.970 --> 00:12:07.163
742 | a relatively slow. These days, it actually
743 | takes a few microseconds on a modern
744 |
745 | 152
746 | 00:12:07.163 --> 00:12:11.259
747 | computer. But still, microseconds for a,
748 | for a, say a four gigahertz processor is
749 |
750 | 153
751 | 00:12:11.259 --> 00:12:15.092
752 | quite a bit of work. And so just keep in
753 | mind that all the exponentiation
754 |
755 | 154
756 | 00:12:15.092 --> 00:12:19.556
757 | operations we talked about. For example,
758 | for determining if a number is a quadratic
759 |
760 | 155
761 | 00:12:19.556 --> 00:12:23.600
762 | residue or not, All those, all those
763 | exponentiations basically take cubic time.
764 |
765 | 156
766 | 00:12:24.780 --> 00:12:29.677
767 | Okay, so that completes our discussion of
768 | arithmetic algorithms, and then in the
769 |
770 | 157
771 | 00:12:29.677 --> 00:12:34.078
772 | next segment we'll start talking about
773 | hard problems, modulo, primes and composites.
--------------------------------------------------------------------------------
/tests/test_main.py:
--------------------------------------------------------------------------------
1 | import tempfile
2 | from src.main import CommandLineArgRunner
3 | from unittest import mock, TestCase
4 | from parameterized import parameterized
5 | from tests.utils.pdf_snapshot import assert_match_pdf_snapshot
6 |
7 |
8 | @mock.patch("time.time", mock.MagicMock(return_value=12345))
9 | class MainTests(TestCase):
10 | @parameterized.expand(
11 | [
12 | ("input_1.mp4", "subtitles_1.srt", "video_1_with_srt_file.pdf"),
13 | ("input_1.mp4", "subtitles_1.vtt", "video_1_with_webvtt_file.pdf"),
14 | ("input_2.mp4", "subtitles_2.srt", "video_2_with_srt_file.pdf"),
15 | ("input_2.mp4", "subtitles_2.vtt", "video_2_with_webvtt_file.pdf"),
16 | ("input_7.mp4", "subtitles_7.srt", "video_7_with_srt_file.pdf"),
17 | ("input_8.mp4", "subtitles_8.srt", "video_8_with_srt_file.pdf"),
18 | ("input_8.mp4", "subtitles_8.vtt", "video_8_with_webvtt_file.pdf"),
19 | ]
20 | )
21 | def test_run_given_video_and_subtitles_should_generate_pdf_correctly(
22 | self, video_filename, subtitle_filename, output_filename
23 | ):
24 | test_output_file = tempfile.NamedTemporaryFile(suffix=".pdf")
25 |
26 | cli = CommandLineArgRunner()
27 | cli.run(
28 | [
29 | f"tests/videos/{video_filename}",
30 | "-s",
31 | f"tests/subtitles/{subtitle_filename}",
32 | "-o",
33 | test_output_file.name,
34 | ]
35 | )
36 |
37 | assert_match_pdf_snapshot(
38 | f"tests/snapshots/snap_test_main/{output_filename}",
39 | test_output_file.name,
40 | )
41 |
42 | @parameterized.expand(
43 | [
44 | ("input_1.mp4", "video_1_without_subtitles.pdf"),
45 | ("input_2.mp4", "video_2_without_subtitles.pdf"),
46 | ("input_7.mp4", "video_7_without_subtitles.pdf"),
47 | ("input_8.mp4", "video_8_without_subtitles.pdf"),
48 | ]
49 | )
50 | def test_run_given_video_and_skip_subtitles_flag_should_generate_pdf_correctly(
51 | self, video_filename, output_filename
52 | ):
53 | test_output_file = tempfile.NamedTemporaryFile(suffix=".pdf")
54 |
55 | cli = CommandLineArgRunner()
56 | cli.run(
57 | [
58 | f"tests/videos/{video_filename}",
59 | "-S",
60 | "-o",
61 | test_output_file.name,
62 | ]
63 | )
64 |
65 | assert_match_pdf_snapshot(
66 | f"tests/snapshots/snap_test_main/{output_filename}",
67 | test_output_file.name,
68 | )
69 |
70 | def test_run_given_video_and_subtitles_and_skip_subtitles_flag_should_throw_error(
71 | self,
72 | ):
73 | test_output_file = tempfile.NamedTemporaryFile(suffix=".pdf")
74 |
75 | cli = CommandLineArgRunner()
76 | func_to_test = lambda: cli.run(
77 | [
78 | "tests/videos/input_1.mp4",
79 | "-S",
80 | "-s",
81 | "tests/subtitles/subtitles_1.srt",
82 | "-o",
83 | test_output_file.name,
84 | ]
85 | )
86 |
87 | self.assertRaises(AssertionError, func_to_test)
88 |
--------------------------------------------------------------------------------
/tests/test_subtitle_segment_finder.py:
--------------------------------------------------------------------------------
1 | import unittest
2 | import snapshottest
3 | from src.time_utils import convert_clock_time_to_timestamp_ms as get_timestamp
4 | from src.subtitle_segment_finder import SubtitleSegmentFinder
5 | from src.subtitle_part import SubtitlePart
6 | from src.subtitle_webvtt_parser import SubtitleWebVTTParser
7 | from src.subtitle_srt_parser import SubtitleSRTParser
8 |
9 |
10 | class SubtitleSplitterTests(snapshottest.TestCase):
11 | def test_get_pages_given_subtitle_without_periods_and_future_timestamp_should_return_correct_pages(
12 | self,
13 | ):
14 | segments = [
15 | SubtitlePart(
16 | get_timestamp("00:00:00"),
17 | get_timestamp("00:00:10"),
18 | "hi there my name is Bob",
19 | )
20 | ]
21 | pager = SubtitleSegmentFinder(segments)
22 | time_breaks = [get_timestamp("00:00:14")]
23 | transcript_pages = pager.get_subtitle_segments(time_breaks)
24 |
25 | self.assertEqual(transcript_pages[0], "hi there my name is Bob")
26 |
27 | def test_get_pages_given_subtitle_without_periods_and_middle_timestamp_should_return_correct_pages(
28 | self,
29 | ):
30 | segments = [
31 | SubtitlePart(
32 | get_timestamp("00:00:00"),
33 | get_timestamp("00:00:10"),
34 | "hi there my name is Bob",
35 | )
36 | ]
37 | pager = SubtitleSegmentFinder(segments)
38 | time_breaks = [get_timestamp("00:00:05")]
39 | transcript_pages = pager.get_subtitle_segments(time_breaks)
40 |
41 | self.assertEqual(transcript_pages[0], "hi there my")
42 |
43 | def test_get_pages_given_subtitle_without_periods_and_past_timestamp_should_return_correct_pages(
44 | self,
45 | ):
46 | segments = [
47 | SubtitlePart(
48 | get_timestamp("00:00:10"),
49 | get_timestamp("00:00:20"),
50 | "hi there my name is Bob",
51 | )
52 | ]
53 | pager = SubtitleSegmentFinder(segments)
54 | time_breaks = [get_timestamp("00:00:05")]
55 | transcript_pages = pager.get_subtitle_segments(time_breaks)
56 |
57 | self.assertEqual(transcript_pages[0], "")
58 |
59 | def test_get_pages_given_subtitle_with_period_should_return_correct_pages(self):
60 | segments = [
61 | SubtitlePart(
62 | get_timestamp("00:00:00"),
63 | get_timestamp("00:00:10"),
64 | "Hi. My name is Bob",
65 | )
66 | ]
67 | pager = SubtitleSegmentFinder(segments)
68 | time_breaks = [get_timestamp("00:00:05")]
69 | transcript_pages = pager.get_subtitle_segments(time_breaks)
70 | self.assertEqual(transcript_pages[0], "Hi.")
71 |
72 | def test_get_pages_given_multiple_subtitle_segments_should_return_correct_pages(
73 | self,
74 | ):
75 | segments = [
76 | SubtitlePart(
77 | get_timestamp("00:00:00"),
78 | get_timestamp("00:00:10"),
79 | "Hi. My name is Bob",
80 | ),
81 | SubtitlePart(
82 | get_timestamp("00:00:10"),
83 | get_timestamp("00:00:20"),
84 | "and his name is Alice. Today, we are",
85 | ),
86 | ]
87 | pager = SubtitleSegmentFinder(segments)
88 | time_breaks = [get_timestamp("00:00:02"), get_timestamp("00:00:15")]
89 | transcript_pages = pager.get_subtitle_segments(time_breaks)
90 |
91 | self.assertEqual(transcript_pages[0], "Hi.")
92 | self.assertEqual(transcript_pages[1], "My name is Bob and his name is Alice.")
93 |
94 | def test_get_pages_given_multiple_subtitle_segments_when_selecting_subtitle_at_break_should_return_correct_pages(
95 | self,
96 | ):
97 | segments = [
98 | SubtitlePart(
99 | get_timestamp("00:00:00"),
100 | get_timestamp("00:00:10"),
101 | "Hi. My name is Bob",
102 | ),
103 | SubtitlePart(
104 | get_timestamp("00:00:10"),
105 | get_timestamp("00:00:20"),
106 | "and his name is Alice. Today, we are",
107 | ),
108 | ]
109 | pager = SubtitleSegmentFinder(segments)
110 | time_breaks = [get_timestamp("00:00:10"), get_timestamp("00:00:20")]
111 | transcript_pages = pager.get_subtitle_segments(time_breaks)
112 |
113 | self.assertEqual(len(transcript_pages), 2)
114 | self.assertEqual(transcript_pages[0], "Hi.")
115 | self.assertEqual(
116 | transcript_pages[1], "My name is Bob and his name is Alice. Today, we are"
117 | )
118 |
119 | def test_get_pages_given_subtitle_1_should_return_correct_pages(self):
120 | segments = SubtitleWebVTTParser(
121 | "tests/subtitles/subtitles_1.vtt"
122 | ).get_subtitle_parts()
123 | pager = SubtitleSegmentFinder(segments)
124 |
125 | breaks = [
126 | get_timestamp("00:00:14"),
127 | get_timestamp("00:02:10"),
128 | get_timestamp("00:05:38"),
129 | get_timestamp("00:07:58"),
130 | get_timestamp("00:10:37"),
131 | get_timestamp("00:10:52"),
132 | get_timestamp("00:10:54"),
133 | ]
134 | transcript_pages = pager.get_subtitle_segments(breaks)
135 |
136 | self.assertEqual(len(transcript_pages), 7)
137 | self.assertMatchSnapshot(transcript_pages)
138 |
139 | def test_get_pages_given_subtitle_8_should_return_correct_pages(self):
140 | segments = SubtitleSRTParser(
141 | "tests/subtitles/subtitles_8.srt"
142 | ).get_subtitle_parts()
143 | pager = SubtitleSegmentFinder(segments)
144 |
145 | breaks = [
146 | get_timestamp("00:00:04"),
147 | get_timestamp("00:00:05"),
148 | get_timestamp("00:00:06"),
149 | get_timestamp("00:00:07"),
150 | get_timestamp("00:00:08"),
151 | get_timestamp("00:00:09"),
152 | get_timestamp("00:00:15"),
153 | get_timestamp("00:00:23"),
154 | get_timestamp("00:00:26"),
155 | get_timestamp("00:00:27"),
156 | get_timestamp("00:00:31"),
157 | get_timestamp("00:00:41"),
158 | get_timestamp("00:00:45"),
159 | get_timestamp("00:00:53"),
160 | get_timestamp("00:01:03"),
161 | get_timestamp("00:01:25"),
162 | get_timestamp("00:01:47"),
163 | get_timestamp("00:04:00"),
164 | get_timestamp("00:04:30"),
165 | get_timestamp("00:05:15"),
166 | get_timestamp("00:05:58"),
167 | get_timestamp("00:07:33"),
168 | get_timestamp("00:08:04"),
169 | get_timestamp("00:08:24"),
170 | get_timestamp("00:09:02"),
171 | get_timestamp("00:09:42"),
172 | get_timestamp("00:10:00"),
173 | ]
174 | transcript_pages = pager.get_subtitle_segments(breaks)
175 |
176 | self.assertEqual(len(transcript_pages), 27)
177 | self.assertMatchSnapshot(transcript_pages)
178 |
179 | def test_get_pages_given_subtitles_with_no_dots_should_return_correct_pages(self):
180 | segments = [
181 | SubtitlePart(
182 | get_timestamp("00:00:00"),
183 | get_timestamp("00:00:10"),
184 | "Hi my name is Bob",
185 | ),
186 | SubtitlePart(
187 | get_timestamp("00:00:10"),
188 | get_timestamp("00:00:20"),
189 | "and his name is Alice Today, we are",
190 | ),
191 | ]
192 | pager = SubtitleSegmentFinder(segments)
193 | time_breaks = [get_timestamp("00:00:08"), get_timestamp("00:00:20")]
194 | transcript_pages = pager.get_subtitle_segments(time_breaks)
195 |
196 | self.assertEqual(len(transcript_pages), 2)
197 | self.assertEqual(transcript_pages[0], "Hi my name is")
198 | self.assertEqual(transcript_pages[1], "Bob and his name is Alice Today, we are")
199 |
--------------------------------------------------------------------------------
/tests/test_subtitle_srt_parser.py:
--------------------------------------------------------------------------------
1 | import tempfile
2 | import unittest
3 | from src.time_utils import convert_clock_time_to_timestamp_ms as get_timestamp
4 | from src.subtitle_srt_parser import SubtitleSRTParser
5 |
6 |
7 | class SubtitleSRTParserTests(unittest.TestCase):
8 | def test_get_subtitle_parts_given_subtitles_should_return_correct_subtitles(self):
9 | with tempfile.NamedTemporaryFile(mode="w+", encoding="utf-8") as tmpFile:
10 | tmpFile.writelines(
11 | [
12 | "1\n",
13 | "00:00:00,000 --> 00:00:04,134\n",
14 | "So now that we understand what a secure\n",
15 | "PRG is, and we understand what semantic\n",
16 | "\n",
17 | "2\n",
18 | "00:00:04,134 --> 00:00:08,425\n",
19 | "security means, we can actually argue that\n",
20 | "a stream cipher with a secure PRG is, in\n",
21 | ]
22 | )
23 | tmpFile.flush()
24 |
25 | segments = SubtitleSRTParser(tmpFile.name).get_subtitle_parts()
26 |
27 | self.assertEqual(len(segments), 2)
28 | self.assertEqual(segments[0].start_time, get_timestamp("00:00:00"))
29 | self.assertEqual(segments[0].end_time, get_timestamp("00:00:04.134"))
30 | self.assertEqual(
31 | segments[0].text,
32 | "So now that we understand what a secure PRG is, and we understand what semantic",
33 | )
34 |
35 | self.assertEqual(segments[1].start_time, get_timestamp("00:00:04.134"))
36 | self.assertEqual(segments[1].end_time, get_timestamp("00:00:08.425"))
37 | self.assertEqual(
38 | segments[1].text,
39 | "security means, we can actually argue that a stream cipher with a secure PRG is, in",
40 | )
41 |
42 | def test_get_subtitle_parts_given_subtitles_with_no_text_should_not_include_empty_subtitles(
43 | self,
44 | ):
45 | with tempfile.NamedTemporaryFile(mode="w+", encoding="utf-8") as tmpFile:
46 | tmpFile.writelines(
47 | [
48 | "1\n",
49 | "00:00:00,000 --> 00:00:04,134\n",
50 | "So now that we understand what a secure\n",
51 | "PRG is, and we understand what semantic\n",
52 | "\n",
53 | "2\n",
54 | "00:00:04,134 --> 00:00:10,134\n",
55 | "\n",
56 | "\n",
57 | "\n",
58 | "3\n",
59 | "00:00:10,134 --> 00:00:18,425\n",
60 | "security means, we can actually argue that\n",
61 | "a stream cipher with a secure PRG is, in\n",
62 | ]
63 | )
64 | tmpFile.flush()
65 |
66 | segments = SubtitleSRTParser(tmpFile.name).get_subtitle_parts()
67 |
68 | self.assertEqual(len(segments), 2)
69 | self.assertEqual(segments[0].start_time, get_timestamp("00:00:00"))
70 | self.assertEqual(segments[0].end_time, get_timestamp("00:00:10.134"))
71 | self.assertEqual(
72 | segments[0].text,
73 | "So now that we understand what a secure PRG is, and we understand what semantic",
74 | )
75 |
76 | self.assertEqual(segments[1].start_time, get_timestamp("00:00:10.134"))
77 | self.assertEqual(segments[1].end_time, get_timestamp("00:00:18.425"))
78 | self.assertEqual(
79 | segments[1].text,
80 | "security means, we can actually argue that a stream cipher with a secure PRG is, in",
81 | )
82 |
83 | def test_get_subtitle_parts_given_subtitles_that_are_not_continuous_should_return_correct_subtitles(
84 | self,
85 | ):
86 | with tempfile.NamedTemporaryFile(mode="w+", encoding="utf-8") as tmpFile:
87 | tmpFile.writelines(
88 | [
89 | "1\n",
90 | "00:00:00,000 --> 00:00:04,134\n",
91 | "So now that we understand what a secure\n",
92 | "PRG is, and we understand what semantic\n",
93 | "\n",
94 | "2\n",
95 | "00:00:04,134 --> 00:00:08,425\n",
96 | "security means, we can actually argue that\n",
97 | "a stream cipher with a secure PRG is, in\n",
98 | ]
99 | )
100 | tmpFile.flush()
101 |
102 | segments = SubtitleSRTParser(tmpFile.name).get_subtitle_parts()
103 |
104 | self.assertEqual(len(segments), 2)
105 | self.assertEqual(segments[0].start_time, get_timestamp("00:00:00"))
106 | self.assertEqual(segments[0].end_time, get_timestamp("00:00:04.134"))
107 | self.assertEqual(
108 | segments[0].text,
109 | "So now that we understand what a secure PRG is, and we understand what semantic",
110 | )
111 |
112 | self.assertEqual(segments[1].start_time, get_timestamp("00:00:04.134"))
113 | self.assertEqual(segments[1].end_time, get_timestamp("00:00:08.425"))
114 | self.assertEqual(
115 | segments[1].text,
116 | "security means, we can actually argue that a stream cipher with a secure PRG is, in",
117 | )
118 |
--------------------------------------------------------------------------------
/tests/test_subtitle_webvtt_parser.py:
--------------------------------------------------------------------------------
1 | import tempfile
2 | import unittest
3 | from src.time_utils import convert_timestamp_ms_to_clock_time as get_clock_time
4 | from src.time_utils import convert_clock_time_to_timestamp_ms as get_timestamp
5 | from src.subtitle_webvtt_parser import SubtitleWebVTTParser
6 |
7 |
8 | class SubtitleWebVTTParserTests(unittest.TestCase):
9 | def test_get_subtitle_parts_given_subtitles_should_return_correct_subtitles(self):
10 | with tempfile.NamedTemporaryFile(mode="w+", encoding="utf-8") as tmpFile:
11 | tmpFile.writelines(
12 | [
13 | "WEBVTT\n",
14 | "\n",
15 | "1\n",
16 | "00:00:00.000 --> 00:00:04.134\n",
17 | "So now that we understand what a secure\n",
18 | "PRG is, and we understand what semantic\n",
19 | "\n",
20 | "2\n",
21 | "00:00:04.134 --> 00:00:08.425\n",
22 | "security means, we can actually argue that\n",
23 | "a stream cipher with a secure PRG is, in\n",
24 | ]
25 | )
26 | tmpFile.flush()
27 |
28 | segments = SubtitleWebVTTParser(tmpFile.name).get_subtitle_parts()
29 |
30 | self.assertEqual(len(segments), 2)
31 | self.assertEqual(segments[0].start_time, get_timestamp("00:00:00"))
32 | self.assertEqual(segments[0].end_time, get_timestamp("00:00:04.134"))
33 | self.assertEqual(
34 | segments[0].text,
35 | "So now that we understand what a secure PRG is, and we understand what semantic",
36 | )
37 |
38 | self.assertEqual(segments[1].start_time, get_timestamp("00:00:04.134"))
39 | self.assertEqual(segments[1].end_time, get_timestamp("00:00:08.425"))
40 | self.assertEqual(
41 | segments[1].text,
42 | "security means, we can actually argue that a stream cipher with a secure PRG is, in",
43 | )
44 |
45 | def test_get_subtitle_parts_given_subtitles_with_no_text_should_not_include_empty_subtitles(
46 | self,
47 | ):
48 | with tempfile.NamedTemporaryFile(mode="w+", encoding="utf-8") as tmpFile:
49 | tmpFile.writelines(
50 | [
51 | "WEBVTT\n",
52 | "\n",
53 | "1\n",
54 | "00:00:00.000 --> 00:00:04.134\n",
55 | "So now that we understand what a secure\n",
56 | "PRG is, and we understand what semantic\n",
57 | "\n",
58 | "2\n",
59 | "00:00:04.134 --> 00:00:10.134\n",
60 | "\n",
61 | "\n",
62 | "\n",
63 | "3\n",
64 | "00:00:10.134 --> 00:00:18.425\n",
65 | "security means, we can actually argue that\n",
66 | "a stream cipher with a secure PRG is, in\n",
67 | ]
68 | )
69 | tmpFile.flush()
70 |
71 | segments = SubtitleWebVTTParser(tmpFile.name).get_subtitle_parts()
72 |
73 | self.assertEqual(len(segments), 2)
74 | self.assertEqual(segments[0].start_time, get_timestamp("00:00:00"))
75 | self.assertEqual(segments[0].end_time, get_timestamp("00:00:10.134"))
76 | self.assertEqual(
77 | segments[0].text,
78 | "So now that we understand what a secure PRG is, and we understand what semantic",
79 | )
80 |
81 | self.assertEqual(segments[1].start_time, get_timestamp("00:00:10.134"))
82 | self.assertEqual(segments[1].end_time, get_timestamp("00:00:18.425"))
83 | self.assertEqual(
84 | segments[1].text,
85 | "security means, we can actually argue that a stream cipher with a secure PRG is, in",
86 | )
87 |
88 | def test_get_subtitle_parts_given_subtitles_that_are_not_continuous_should_return_correct_subtitles(
89 | self,
90 | ):
91 | with tempfile.NamedTemporaryFile(mode="w+", encoding="utf-8") as tmpFile:
92 | tmpFile.writelines(
93 | [
94 | "WEBVTT\n",
95 | "\n",
96 | "1\n",
97 | "00:00:00.000 --> 00:00:04.134\n",
98 | "So now that we understand what a secure\n",
99 | "PRG is, and we understand what semantic\n",
100 | "\n",
101 | "2\n",
102 | "00:00:05.000 --> 00:00:08.425\n",
103 | "security means, we can actually argue that\n",
104 | "a stream cipher with a secure PRG is, in\n",
105 | ]
106 | )
107 | tmpFile.flush()
108 |
109 | segments = SubtitleWebVTTParser(tmpFile.name).get_subtitle_parts()
110 |
111 | self.assertEqual(len(segments), 2)
112 | self.assertEqual(segments[0].start_time, get_timestamp("00:00:00"))
113 | self.assertEqual(segments[0].end_time, get_timestamp("00:00:05.000"))
114 | self.assertEqual(
115 | segments[0].text,
116 | "So now that we understand what a secure PRG is, and we understand what semantic",
117 | )
118 |
119 | self.assertEqual(segments[1].start_time, get_timestamp("00:00:05.000"))
120 | self.assertEqual(segments[1].end_time, get_timestamp("00:00:08.425"))
121 | self.assertEqual(
122 | segments[1].text,
123 | "security means, we can actually argue that a stream cipher with a secure PRG is, in",
124 | )
125 |
--------------------------------------------------------------------------------
/tests/test_time_utils.py:
--------------------------------------------------------------------------------
1 | import unittest
2 | from src.time_utils import convert_timestamp_ms_to_clock_time as get_clock_time
3 | from src.time_utils import convert_clock_time_to_timestamp_ms as get_timestamp
4 |
5 |
6 | class TimeUtilsTests(unittest.TestCase):
7 | def test_given_hh_mm_ss_it_should_convert_to_clock_time_correctly(self):
8 | self.assertEqual(get_clock_time(7538000), "02:05:38")
9 |
10 | def test_given_mm_ss_it_should_convert_to_clock_time_correctly(self):
11 | self.assertEqual(get_clock_time(338000), "00:05:38")
12 |
13 | def test_given_seconds_it_should_convert_to_clock_time_correctly(self):
14 | self.assertEqual(get_clock_time(38456), "00:00:38.456")
15 |
16 | def test_given_zero_seconds_it_should_convert_to_clock_time_correctly(self):
17 | self.assertEqual(get_clock_time(456), "00:00:00.456")
18 |
19 | def test_given_timestamp_ms_with_hours_it_should_convert_to_timestamp_ms_correctly(
20 | self,
21 | ):
22 | self.assertEqual(get_timestamp("02:05:38"), 7538000)
23 |
24 | def test_given_time_without_hours_it_should_convert_to_timestamp_ms_correctly(self):
25 | self.assertEqual(get_timestamp("00:05:38"), 338000)
26 |
27 | def test_given_time_minutes_or_hours_it_should_convert_to_timestamp_ms_correctly(
28 | self,
29 | ):
30 | self.assertEqual(get_timestamp("00:00:38.456"), 38456)
31 |
32 | def test_given_zero_seconds_it_should_convert_to_timestamp_ms_correctly(self):
33 | self.assertEqual(get_timestamp("00:00:00.456"), 456)
34 |
--------------------------------------------------------------------------------
/tests/test_video_segment_finder.py:
--------------------------------------------------------------------------------
1 | import unittest
2 | from src.video_segment_finder import VideoSegmentFinder # get_frames
3 | from src.time_utils import convert_timestamp_ms_to_clock_time as get_clock
4 |
5 |
6 | class VideoBreaksTest(unittest.TestCase):
7 | def test_get_frames_of_video_with_human_should_return_correct_breaks(self):
8 | data = VideoSegmentFinder().get_best_segment_frames("tests/videos/input_1.mp4")
9 | # data = get_frames('videos/input_1.mp4')
10 | frame_nums = sorted(data.keys())
11 |
12 | self.assertEqual(len(frame_nums), 7)
13 |
14 | # Check if the timestamps (in ms) matches the ones that we picked out manually
15 | self.assertEqual(
16 | get_clock(data[frame_nums[0]]["timestamp"]), "00:00:14.147480814147482"
17 | )
18 | self.assertEqual(
19 | get_clock(data[frame_nums[1]]["timestamp"]), "00:02:10.330330330330325"
20 | )
21 | self.assertEqual(
22 | get_clock(data[frame_nums[2]]["timestamp"]), "00:05:38.071404738071436"
23 | )
24 | self.assertEqual(
25 | get_clock(data[frame_nums[3]]["timestamp"]), "00:07:58.41174507841177"
26 | )
27 | self.assertEqual(
28 | get_clock(data[frame_nums[4]]["timestamp"]), "00:10:37.53753753753752"
29 | )
30 | self.assertEqual(
31 | get_clock(data[frame_nums[5]]["timestamp"]), "00:10:52.05205205205211"
32 | )
33 | self.assertEqual(
34 | get_clock(data[frame_nums[6]]["timestamp"]), "00:10:54.38772105438775"
35 | )
36 |
37 | def test_get_frames_of_video_without_human_should_return_correct_breaks(self):
38 | data = VideoSegmentFinder().get_best_segment_frames("tests/videos/input_2.mp4")
39 | frame_nums = sorted(data.keys())
40 |
41 | self.assertEqual(len(frame_nums), 7)
42 |
43 | # Check if the timestamps (in ms) matches the ones that we picked out manually
44 | self.assertEqual(
45 | get_clock(data[frame_nums[0]]["timestamp"]), "00:00:04.337671004337671"
46 | )
47 | self.assertEqual(
48 | get_clock(data[frame_nums[1]]["timestamp"]), "00:00:49.215882549215884"
49 | )
50 | self.assertEqual(
51 | get_clock(data[frame_nums[2]]["timestamp"]), "00:04:10.784117450784136"
52 | )
53 | self.assertEqual(
54 | get_clock(data[frame_nums[3]]["timestamp"]), "00:07:59.01234567901236"
55 | )
56 | self.assertEqual(
57 | get_clock(data[frame_nums[4]]["timestamp"]), "00:11:09.669669669669704"
58 | )
59 | self.assertEqual(
60 | get_clock(data[frame_nums[5]]["timestamp"]), "00:12:24.878211544878198"
61 | )
62 | self.assertEqual(
63 | get_clock(data[frame_nums[6]]["timestamp"]), "00:12:34.788121454788254"
64 | )
65 |
66 | def test_get_frames_of_video_with_lots_of_animations_should_return_correct_breaks(
67 | self,
68 | ):
69 | data = VideoSegmentFinder().get_best_segment_frames("tests/videos/input_3.mp4")
70 | frame_nums = sorted(data.keys())
71 |
72 | self.assertEqual(len(frame_nums), 12)
73 |
74 | # Check if the timestamps (in ms) matches the ones that we picked out manually
75 | self.assertEqual(
76 | get_clock(data[frame_nums[0]]["timestamp"]), "00:00:37.69527777777778"
77 | )
78 | self.assertEqual(
79 | get_clock(data[frame_nums[1]]["timestamp"]), "00:00:56.49292222222223"
80 | )
81 | self.assertEqual(
82 | get_clock(data[frame_nums[2]]["timestamp"]), "00:01:13.5407888888889"
83 | )
84 | self.assertEqual(
85 | get_clock(data[frame_nums[3]]["timestamp"]), "00:01:43.53703333333334"
86 | )
87 | self.assertEqual(
88 | get_clock(data[frame_nums[4]]["timestamp"]), "00:02:57.783333333333346"
89 | )
90 | self.assertEqual(
91 | get_clock(data[frame_nums[5]]["timestamp"]), "00:03:33.183333333333344"
92 | )
93 | self.assertEqual(
94 | get_clock(data[frame_nums[6]]["timestamp"]), "00:03:45.98333333333334"
95 | )
96 | self.assertEqual(
97 | get_clock(data[frame_nums[7]]["timestamp"]), "00:03:59.183333333333344"
98 | )
99 | self.assertEqual(
100 | get_clock(data[frame_nums[8]]["timestamp"]), "00:04:09.833333333333343"
101 | )
102 | self.assertEqual(
103 | get_clock(data[frame_nums[9]]["timestamp"]), "00:04:42.48333333333337"
104 | )
105 | self.assertEqual(
106 | get_clock(data[frame_nums[10]]["timestamp"]), "00:05:52.933333333333316"
107 | )
108 | self.assertEqual(
109 | get_clock(data[frame_nums[11]]["timestamp"]), "00:06:40.98333333333337"
110 | )
111 |
112 | def test_get_frames_of_video_with_blur_animations_should_return_correct_breaks(
113 | self,
114 | ):
115 | data = VideoSegmentFinder().get_best_segment_frames("tests/videos/input_4.mp4")
116 | frame_nums = sorted(data.keys())
117 |
118 | self.assertEqual(len(frame_nums), 2)
119 |
120 | # Check if the timestamps (in ms) matches the ones that we picked out manually
121 | self.assertEqual(
122 | get_clock(data[frame_nums[0]]["timestamp"]), "00:00:03.7495333333333334"
123 | )
124 | self.assertEqual(
125 | get_clock(data[frame_nums[1]]["timestamp"]), "00:00:07.649044444444445"
126 | )
127 |
128 | def test_get_frames_of_video_with_blur_annotations_should_return_correct_breaks(
129 | self,
130 | ):
131 | data = VideoSegmentFinder().get_best_segment_frames("tests/videos/input_5.mp4")
132 | frame_nums = sorted(data.keys())
133 |
134 | self.assertEqual(len(frame_nums), 2)
135 |
136 | # Check if the timestamps (in ms) matches the ones that we picked out manually
137 | self.assertEqual(
138 | get_clock(data[frame_nums[0]]["timestamp"]), "00:01:54.34768101434769"
139 | )
140 | self.assertEqual(
141 | get_clock(data[frame_nums[1]]["timestamp"]), "00:02:25.47881214547882"
142 | )
143 |
144 | def test_get_frames_of_video_with_add_animations_should_return_correct_breaks(self):
145 | data = VideoSegmentFinder().get_best_segment_frames("tests/videos/input_6.mp4")
146 | frame_nums = sorted(data.keys())
147 |
148 | self.assertEqual(len(frame_nums), 3)
149 |
150 | # Check if the timestamps (in ms) matches the ones that we picked out manually
151 | self.assertEqual(get_clock(data[frame_nums[0]]["timestamp"]), "00:01:31.15")
152 | self.assertEqual(
153 | get_clock(data[frame_nums[1]]["timestamp"]), "00:01:34.850000000000016"
154 | )
155 | self.assertEqual(get_clock(data[frame_nums[2]]["timestamp"]), "00:01:41.5")
156 |
--------------------------------------------------------------------------------
/tests/utils/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/utils/__init__.py
--------------------------------------------------------------------------------
/tests/utils/pdf_snapshot.py:
--------------------------------------------------------------------------------
1 | import tempfile
2 | import os
3 | from os import path
4 | import shutil
5 | import pypdftk
6 | from wand.image import Image
7 |
8 |
9 | def assert_match_pdf_snapshot(expected_filepath, actual_filepath, max_diff_val=0.0):
10 | """Verifies if the pdf matches the expected pdf snapshot
11 | If the pdf file doesn't exist yet, it will save it
12 |
13 | Parameters
14 | ----------
15 | expected_filepath : str
16 | The file path to the expected pdf file
17 | actual_filepath: str
18 | The file path to the actual pdf file (must be different than expected_filepath)
19 | max_diff_val: num
20 | The max. value the difference between any pages, from 0-100
21 | """
22 |
23 | if not path.exists(actual_filepath):
24 | raise ValueError(f"Actual file {actual_filepath} doesn't exist")
25 |
26 | if not path.exists(expected_filepath):
27 | os.makedirs(os.path.dirname(expected_filepath), exist_ok=True)
28 | shutil.copy(actual_filepath, expected_filepath)
29 | else:
30 | # Get num pages for each pdf
31 | expected_num_pages = pypdftk.get_num_pages(expected_filepath)
32 | actual_num_pages = pypdftk.get_num_pages(actual_filepath)
33 |
34 | if expected_num_pages != actual_num_pages:
35 | raise ValueError(
36 | f"Expected {expected_num_pages} num pages; got {actual_num_pages} pages"
37 | )
38 |
39 | # Create temp dirs for each pdf
40 | try:
41 | expected_split_dir = tempfile.TemporaryDirectory()
42 | actual_split_dir = tempfile.TemporaryDirectory()
43 |
44 | # Split the pdf into different pages
45 | exp_split_files = pypdftk.split(expected_filepath, expected_split_dir.name)[
46 | 1:
47 | ]
48 | act_split_files = pypdftk.split(actual_filepath, actual_split_dir.name)[1:]
49 |
50 | # Compare each page
51 | has_diff = False
52 | for page_idx in range(len(exp_split_files)):
53 | with Image(filename=exp_split_files[page_idx]) as expected_page:
54 | with Image(filename=act_split_files[page_idx]) as actual_page:
55 | expected_page.fuzz = 0.25 * expected_page.quantum_range
56 | expected_page.artifacts["compare:highlight-color"] = "red"
57 | expected_page.artifacts[
58 | "compare:lowlight-color"
59 | ] = "transparent"
60 | diff_img, diff_val = expected_page.compare(
61 | actual_page, "root_mean_square"
62 | )
63 |
64 | if diff_val > max_diff_val:
65 | diff_img_filepath = os.path.join(
66 | os.path.dirname(expected_filepath),
67 | f"{os.path.basename(expected_filepath)}-{page_idx}-diff.jpg",
68 | )
69 | diff_img.save(filename=diff_img_filepath)
70 | has_diff = True
71 |
72 | diff_img.close()
73 |
74 | if has_diff:
75 | print("One or more pages in the pdf is different")
76 | print(f"See folder in {os.path.dirname(expected_filepath)} for diff")
77 | raise ValueError()
78 | finally:
79 | # Clean up the dirs
80 | expected_split_dir.cleanup()
81 | actual_split_dir.cleanup()
82 |
--------------------------------------------------------------------------------
/tests/videos/input_1.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_1.mp4
--------------------------------------------------------------------------------
/tests/videos/input_2.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_2.mp4
--------------------------------------------------------------------------------
/tests/videos/input_3.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_3.mp4
--------------------------------------------------------------------------------
/tests/videos/input_4.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_4.mp4
--------------------------------------------------------------------------------
/tests/videos/input_5.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_5.mp4
--------------------------------------------------------------------------------
/tests/videos/input_6.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_6.mp4
--------------------------------------------------------------------------------
/tests/videos/input_7.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_7.mp4
--------------------------------------------------------------------------------
/tests/videos/input_8.mp4:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/EKarton/Lecture-Video-to-PDF/acad45548ed706d10b0820bb66549690f6a8c16d/tests/videos/input_8.mp4
--------------------------------------------------------------------------------