├── .github
├── FUNDING.yml
└── workflows
│ └── python-publish.yml
├── .gitignore
├── .pre-commit-config.yaml
├── .style.yapf
├── LICENSE
├── README.md
├── README_ch.md
├── docs
├── images
│ ├── andy.png
│ ├── multi.png
│ ├── source.gif
│ ├── swapped.gif
│ └── trump.jpg
└── test
│ ├── condition.jpg
│ ├── dst1.png
│ ├── dst2.png
│ ├── multi.png
│ ├── taitan.jpeg
│ ├── target_pose_reference.jpg
│ └── trump.jpg
├── dofaker
├── __init__.py
├── face_core.py
├── face_det
│ ├── __init__.py
│ └── face_analysis.py
├── face_enhance
│ ├── __init__.py
│ └── gfpgan.py
├── face_swap
│ ├── __init__.py
│ ├── base_swapper.py
│ └── inswapper.py
├── pose
│ ├── __init__.py
│ ├── pose_estimator.py
│ ├── pose_transfer.py
│ └── pose_utils.py
├── pose_core.py
├── super_resolution
│ ├── __init__.py
│ └── bsrgan.py
├── transforms
│ ├── __init__.py
│ └── functional.py
└── utils
│ ├── __init__.py
│ ├── download.py
│ ├── utils.py
│ └── weights_urls.py
├── requirements.txt
├── run_faceswapper.py
├── run_posetransfer.py
├── setup.py
├── test.sh
└── web_ui.py
/.github/FUNDING.yml:
--------------------------------------------------------------------------------
1 | # These are supported funding model platforms
2 |
3 | github: # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
4 | patreon: # Replace with a single Patreon username
5 | open_collective: # Replace with a single Open Collective username
6 | ko_fi: # Replace with a single Ko-fi username
7 | tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
8 | community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
9 | liberapay: # Replace with a single Liberapay username
10 | issuehunt: # Replace with a single IssueHunt username
11 | otechie: # Replace with a single Otechie username
12 | lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
13 | custom: []
14 |
--------------------------------------------------------------------------------
/.github/workflows/python-publish.yml:
--------------------------------------------------------------------------------
1 | # This workflow will upload a Python Package using Twine when a release is created
2 | # For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python#publishing-to-package-registries
3 |
4 | # This workflow uses actions that are not certified by GitHub.
5 | # They are provided by a third-party and are governed by
6 | # separate terms of service, privacy policy, and support
7 | # documentation.
8 |
9 | name: Upload Python Package
10 |
11 | on:
12 | release:
13 | types: [published]
14 |
15 | permissions:
16 | contents: read
17 |
18 | jobs:
19 | deploy:
20 |
21 | runs-on: ubuntu-latest
22 |
23 | steps:
24 | - uses: actions/checkout@v3
25 | - name: Set up Python
26 | uses: actions/setup-python@v3
27 | with:
28 | python-version: '3.x'
29 | - name: Install dependencies
30 | run: |
31 | python -m pip install --upgrade pip
32 | pip install build
33 | - name: Build package
34 | run: python -m build
35 | - name: Publish package
36 | uses: pypa/gh-action-pypi-publish@27b31702a0e7fc50959f5ad993c78deac1bdfc29
37 | with:
38 | user: __token__
39 | password: ${{ secrets.PYPI_API_TOKEN }}
40 |
--------------------------------------------------------------------------------
/.gitignore:
--------------------------------------------------------------------------------
1 | # Byte-compiled / optimized / DLL files
2 | __pycache__/
3 | *.py[cod]
4 | *$py.class
5 | output/
6 | weights/
7 |
8 | # C extensions
9 | *.so
10 |
11 | # Distribution / packaging
12 | .Python
13 | build/
14 | develop-eggs/
15 | dist/
16 | downloads/
17 | eggs/
18 | .eggs/
19 | lib/
20 | lib64/
21 | parts/
22 | sdist/
23 | var/
24 | wheels/
25 | share/python-wheels/
26 | *.egg-info/
27 | .installed.cfg
28 | *.egg
29 | MANIFEST
30 |
31 | # PyInstaller
32 | # Usually these files are written by a python script from a template
33 | # before PyInstaller builds the exe, so as to inject date/other infos into it.
34 | *.manifest
35 | *.spec
36 |
37 | # Installer logs
38 | pip-log.txt
39 | pip-delete-this-directory.txt
40 |
41 | # Unit test / coverage reports
42 | htmlcov/
43 | .tox/
44 | .nox/
45 | .coverage
46 | .coverage.*
47 | .cache
48 | nosetests.xml
49 | coverage.xml
50 | *.cover
51 | *.py,cover
52 | .hypothesis/
53 | .pytest_cache/
54 | cover/
55 |
56 | # Translations
57 | *.mo
58 | *.pot
59 |
60 | # Django stuff:
61 | *.log
62 | local_settings.py
63 | db.sqlite3
64 | db.sqlite3-journal
65 |
66 | # Flask stuff:
67 | instance/
68 | .webassets-cache
69 |
70 | # Scrapy stuff:
71 | .scrapy
72 |
73 | # Sphinx documentation
74 | docs/_build/
75 |
76 | # PyBuilder
77 | .pybuilder/
78 | target/
79 |
80 | # Jupyter Notebook
81 | .ipynb_checkpoints
82 |
83 | # IPython
84 | profile_default/
85 | ipython_config.py
86 |
87 | # pyenv
88 | # For a library or package, you might want to ignore these files since the code is
89 | # intended to run in multiple environments; otherwise, check them in:
90 | # .python-version
91 |
92 | # pipenv
93 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
94 | # However, in case of collaboration, if having platform-specific dependencies or dependencies
95 | # having no cross-platform support, pipenv may install dependencies that don't work, or not
96 | # install all needed dependencies.
97 | #Pipfile.lock
98 |
99 | # poetry
100 | # Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
101 | # This is especially recommended for binary packages to ensure reproducibility, and is more
102 | # commonly ignored for libraries.
103 | # https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
104 | #poetry.lock
105 |
106 | # pdm
107 | # Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
108 | #pdm.lock
109 | # pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
110 | # in version control.
111 | # https://pdm.fming.dev/#use-with-ide
112 | .pdm.toml
113 |
114 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
115 | __pypackages__/
116 |
117 | # Celery stuff
118 | celerybeat-schedule
119 | celerybeat.pid
120 |
121 | # SageMath parsed files
122 | *.sage.py
123 |
124 | # Environments
125 | .env
126 | .venv
127 | env/
128 | venv/
129 | ENV/
130 | env.bak/
131 | venv.bak/
132 |
133 | # Spyder project settings
134 | .spyderproject
135 | .spyproject
136 |
137 | # Rope project settings
138 | .ropeproject
139 |
140 | # mkdocs documentation
141 | /site
142 |
143 | # mypy
144 | .mypy_cache/
145 | .dmypy.json
146 | dmypy.json
147 |
148 | # Pyre type checker
149 | .pyre/
150 |
151 | # pytype static type analyzer
152 | .pytype/
153 |
154 | # Cython debug symbols
155 | cython_debug/
156 |
157 | # PyCharm
158 | # JetBrains specific template is maintained in a separate JetBrains.gitignore that can
159 | # be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
160 | # and can be added to the global gitignore or merged into this file. For a more nuclear
161 | # option (not recommended) you can uncomment the following to ignore the entire idea folder.
162 | #.idea/
163 |
--------------------------------------------------------------------------------
/.pre-commit-config.yaml:
--------------------------------------------------------------------------------
1 | repos:
2 | - repo: local
3 | hooks:
4 | - id: yapf
5 | name: yapf
6 | entry: yapf --style .style.yapf -i
7 | language: system
8 | files: \.py$
9 |
10 | - repo: https://github.com/pre-commit/pre-commit-hooks
11 | rev: a11d9314b22d8f8c7556443875b731ef05965464
12 | hooks:
13 | - id: check-merge-conflict
14 | - id: check-symlinks
15 | - id: end-of-file-fixer
16 | - id: trailing-whitespace
17 | - id: detect-private-key
18 | - id: check-added-large-files
19 |
20 | - repo: local
21 | hooks:
22 | - id: flake8
23 | name: flake8
24 | entry: flake8 --count --select=E9,F63,F7,F82 --show-source --statistics
25 | language: system
26 | files: \.py$
27 |
28 | - repo: local
29 | hooks:
30 | - id: clang-format-with-version-check
31 | name: clang-format
32 | description: Format files with ClangFormat
33 | entry: bash .clang_format.hook -style=Google -i
34 | language: system
35 | files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|cuh|proto)$
36 |
--------------------------------------------------------------------------------
/.style.yapf:
--------------------------------------------------------------------------------
1 | [style]
2 | based_on_style = pep8
3 | column_limit = 80
--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
1 | GNU GENERAL PUBLIC LICENSE
2 | Version 3, 29 June 2007
3 |
4 | Copyright (C) 2007 Free Software Foundation, Inc.
5 | Everyone is permitted to copy and distribute verbatim copies
6 | of this license document, but changing it is not allowed.
7 |
8 | Preamble
9 |
10 | The GNU General Public License is a free, copyleft license for
11 | software and other kinds of works.
12 |
13 | The licenses for most software and other practical works are designed
14 | to take away your freedom to share and change the works. By contrast,
15 | the GNU General Public License is intended to guarantee your freedom to
16 | share and change all versions of a program--to make sure it remains free
17 | software for all its users. We, the Free Software Foundation, use the
18 | GNU General Public License for most of our software; it applies also to
19 | any other work released this way by its authors. You can apply it to
20 | your programs, too.
21 |
22 | When we speak of free software, we are referring to freedom, not
23 | price. Our General Public Licenses are designed to make sure that you
24 | have the freedom to distribute copies of free software (and charge for
25 | them if you wish), that you receive source code or can get it if you
26 | want it, that you can change the software or use pieces of it in new
27 | free programs, and that you know you can do these things.
28 |
29 | To protect your rights, we need to prevent others from denying you
30 | these rights or asking you to surrender the rights. Therefore, you have
31 | certain responsibilities if you distribute copies of the software, or if
32 | you modify it: responsibilities to respect the freedom of others.
33 |
34 | For example, if you distribute copies of such a program, whether
35 | gratis or for a fee, you must pass on to the recipients the same
36 | freedoms that you received. You must make sure that they, too, receive
37 | or can get the source code. And you must show them these terms so they
38 | know their rights.
39 |
40 | Developers that use the GNU GPL protect your rights with two steps:
41 | (1) assert copyright on the software, and (2) offer you this License
42 | giving you legal permission to copy, distribute and/or modify it.
43 |
44 | For the developers' and authors' protection, the GPL clearly explains
45 | that there is no warranty for this free software. For both users' and
46 | authors' sake, the GPL requires that modified versions be marked as
47 | changed, so that their problems will not be attributed erroneously to
48 | authors of previous versions.
49 |
50 | Some devices are designed to deny users access to install or run
51 | modified versions of the software inside them, although the manufacturer
52 | can do so. This is fundamentally incompatible with the aim of
53 | protecting users' freedom to change the software. The systematic
54 | pattern of such abuse occurs in the area of products for individuals to
55 | use, which is precisely where it is most unacceptable. Therefore, we
56 | have designed this version of the GPL to prohibit the practice for those
57 | products. If such problems arise substantially in other domains, we
58 | stand ready to extend this provision to those domains in future versions
59 | of the GPL, as needed to protect the freedom of users.
60 |
61 | Finally, every program is threatened constantly by software patents.
62 | States should not allow patents to restrict development and use of
63 | software on general-purpose computers, but in those that do, we wish to
64 | avoid the special danger that patents applied to a free program could
65 | make it effectively proprietary. To prevent this, the GPL assures that
66 | patents cannot be used to render the program non-free.
67 |
68 | The precise terms and conditions for copying, distribution and
69 | modification follow.
70 |
71 | TERMS AND CONDITIONS
72 |
73 | 0. Definitions.
74 |
75 | "This License" refers to version 3 of the GNU General Public License.
76 |
77 | "Copyright" also means copyright-like laws that apply to other kinds of
78 | works, such as semiconductor masks.
79 |
80 | "The Program" refers to any copyrightable work licensed under this
81 | License. Each licensee is addressed as "you". "Licensees" and
82 | "recipients" may be individuals or organizations.
83 |
84 | To "modify" a work means to copy from or adapt all or part of the work
85 | in a fashion requiring copyright permission, other than the making of an
86 | exact copy. The resulting work is called a "modified version" of the
87 | earlier work or a work "based on" the earlier work.
88 |
89 | A "covered work" means either the unmodified Program or a work based
90 | on the Program.
91 |
92 | To "propagate" a work means to do anything with it that, without
93 | permission, would make you directly or secondarily liable for
94 | infringement under applicable copyright law, except executing it on a
95 | computer or modifying a private copy. Propagation includes copying,
96 | distribution (with or without modification), making available to the
97 | public, and in some countries other activities as well.
98 |
99 | To "convey" a work means any kind of propagation that enables other
100 | parties to make or receive copies. Mere interaction with a user through
101 | a computer network, with no transfer of a copy, is not conveying.
102 |
103 | An interactive user interface displays "Appropriate Legal Notices"
104 | to the extent that it includes a convenient and prominently visible
105 | feature that (1) displays an appropriate copyright notice, and (2)
106 | tells the user that there is no warranty for the work (except to the
107 | extent that warranties are provided), that licensees may convey the
108 | work under this License, and how to view a copy of this License. If
109 | the interface presents a list of user commands or options, such as a
110 | menu, a prominent item in the list meets this criterion.
111 |
112 | 1. Source Code.
113 |
114 | The "source code" for a work means the preferred form of the work
115 | for making modifications to it. "Object code" means any non-source
116 | form of a work.
117 |
118 | A "Standard Interface" means an interface that either is an official
119 | standard defined by a recognized standards body, or, in the case of
120 | interfaces specified for a particular programming language, one that
121 | is widely used among developers working in that language.
122 |
123 | The "System Libraries" of an executable work include anything, other
124 | than the work as a whole, that (a) is included in the normal form of
125 | packaging a Major Component, but which is not part of that Major
126 | Component, and (b) serves only to enable use of the work with that
127 | Major Component, or to implement a Standard Interface for which an
128 | implementation is available to the public in source code form. A
129 | "Major Component", in this context, means a major essential component
130 | (kernel, window system, and so on) of the specific operating system
131 | (if any) on which the executable work runs, or a compiler used to
132 | produce the work, or an object code interpreter used to run it.
133 |
134 | The "Corresponding Source" for a work in object code form means all
135 | the source code needed to generate, install, and (for an executable
136 | work) run the object code and to modify the work, including scripts to
137 | control those activities. However, it does not include the work's
138 | System Libraries, or general-purpose tools or generally available free
139 | programs which are used unmodified in performing those activities but
140 | which are not part of the work. For example, Corresponding Source
141 | includes interface definition files associated with source files for
142 | the work, and the source code for shared libraries and dynamically
143 | linked subprograms that the work is specifically designed to require,
144 | such as by intimate data communication or control flow between those
145 | subprograms and other parts of the work.
146 |
147 | The Corresponding Source need not include anything that users
148 | can regenerate automatically from other parts of the Corresponding
149 | Source.
150 |
151 | The Corresponding Source for a work in source code form is that
152 | same work.
153 |
154 | 2. Basic Permissions.
155 |
156 | All rights granted under this License are granted for the term of
157 | copyright on the Program, and are irrevocable provided the stated
158 | conditions are met. This License explicitly affirms your unlimited
159 | permission to run the unmodified Program. The output from running a
160 | covered work is covered by this License only if the output, given its
161 | content, constitutes a covered work. This License acknowledges your
162 | rights of fair use or other equivalent, as provided by copyright law.
163 |
164 | You may make, run and propagate covered works that you do not
165 | convey, without conditions so long as your license otherwise remains
166 | in force. You may convey covered works to others for the sole purpose
167 | of having them make modifications exclusively for you, or provide you
168 | with facilities for running those works, provided that you comply with
169 | the terms of this License in conveying all material for which you do
170 | not control copyright. Those thus making or running the covered works
171 | for you must do so exclusively on your behalf, under your direction
172 | and control, on terms that prohibit them from making any copies of
173 | your copyrighted material outside their relationship with you.
174 |
175 | Conveying under any other circumstances is permitted solely under
176 | the conditions stated below. Sublicensing is not allowed; section 10
177 | makes it unnecessary.
178 |
179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
180 |
181 | No covered work shall be deemed part of an effective technological
182 | measure under any applicable law fulfilling obligations under article
183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or
184 | similar laws prohibiting or restricting circumvention of such
185 | measures.
186 |
187 | When you convey a covered work, you waive any legal power to forbid
188 | circumvention of technological measures to the extent such circumvention
189 | is effected by exercising rights under this License with respect to
190 | the covered work, and you disclaim any intention to limit operation or
191 | modification of the work as a means of enforcing, against the work's
192 | users, your or third parties' legal rights to forbid circumvention of
193 | technological measures.
194 |
195 | 4. Conveying Verbatim Copies.
196 |
197 | You may convey verbatim copies of the Program's source code as you
198 | receive it, in any medium, provided that you conspicuously and
199 | appropriately publish on each copy an appropriate copyright notice;
200 | keep intact all notices stating that this License and any
201 | non-permissive terms added in accord with section 7 apply to the code;
202 | keep intact all notices of the absence of any warranty; and give all
203 | recipients a copy of this License along with the Program.
204 |
205 | You may charge any price or no price for each copy that you convey,
206 | and you may offer support or warranty protection for a fee.
207 |
208 | 5. Conveying Modified Source Versions.
209 |
210 | You may convey a work based on the Program, or the modifications to
211 | produce it from the Program, in the form of source code under the
212 | terms of section 4, provided that you also meet all of these conditions:
213 |
214 | a) The work must carry prominent notices stating that you modified
215 | it, and giving a relevant date.
216 |
217 | b) The work must carry prominent notices stating that it is
218 | released under this License and any conditions added under section
219 | 7. This requirement modifies the requirement in section 4 to
220 | "keep intact all notices".
221 |
222 | c) You must license the entire work, as a whole, under this
223 | License to anyone who comes into possession of a copy. This
224 | License will therefore apply, along with any applicable section 7
225 | additional terms, to the whole of the work, and all its parts,
226 | regardless of how they are packaged. This License gives no
227 | permission to license the work in any other way, but it does not
228 | invalidate such permission if you have separately received it.
229 |
230 | d) If the work has interactive user interfaces, each must display
231 | Appropriate Legal Notices; however, if the Program has interactive
232 | interfaces that do not display Appropriate Legal Notices, your
233 | work need not make them do so.
234 |
235 | A compilation of a covered work with other separate and independent
236 | works, which are not by their nature extensions of the covered work,
237 | and which are not combined with it such as to form a larger program,
238 | in or on a volume of a storage or distribution medium, is called an
239 | "aggregate" if the compilation and its resulting copyright are not
240 | used to limit the access or legal rights of the compilation's users
241 | beyond what the individual works permit. Inclusion of a covered work
242 | in an aggregate does not cause this License to apply to the other
243 | parts of the aggregate.
244 |
245 | 6. Conveying Non-Source Forms.
246 |
247 | You may convey a covered work in object code form under the terms
248 | of sections 4 and 5, provided that you also convey the
249 | machine-readable Corresponding Source under the terms of this License,
250 | in one of these ways:
251 |
252 | a) Convey the object code in, or embodied in, a physical product
253 | (including a physical distribution medium), accompanied by the
254 | Corresponding Source fixed on a durable physical medium
255 | customarily used for software interchange.
256 |
257 | b) Convey the object code in, or embodied in, a physical product
258 | (including a physical distribution medium), accompanied by a
259 | written offer, valid for at least three years and valid for as
260 | long as you offer spare parts or customer support for that product
261 | model, to give anyone who possesses the object code either (1) a
262 | copy of the Corresponding Source for all the software in the
263 | product that is covered by this License, on a durable physical
264 | medium customarily used for software interchange, for a price no
265 | more than your reasonable cost of physically performing this
266 | conveying of source, or (2) access to copy the
267 | Corresponding Source from a network server at no charge.
268 |
269 | c) Convey individual copies of the object code with a copy of the
270 | written offer to provide the Corresponding Source. This
271 | alternative is allowed only occasionally and noncommercially, and
272 | only if you received the object code with such an offer, in accord
273 | with subsection 6b.
274 |
275 | d) Convey the object code by offering access from a designated
276 | place (gratis or for a charge), and offer equivalent access to the
277 | Corresponding Source in the same way through the same place at no
278 | further charge. You need not require recipients to copy the
279 | Corresponding Source along with the object code. If the place to
280 | copy the object code is a network server, the Corresponding Source
281 | may be on a different server (operated by you or a third party)
282 | that supports equivalent copying facilities, provided you maintain
283 | clear directions next to the object code saying where to find the
284 | Corresponding Source. Regardless of what server hosts the
285 | Corresponding Source, you remain obligated to ensure that it is
286 | available for as long as needed to satisfy these requirements.
287 |
288 | e) Convey the object code using peer-to-peer transmission, provided
289 | you inform other peers where the object code and Corresponding
290 | Source of the work are being offered to the general public at no
291 | charge under subsection 6d.
292 |
293 | A separable portion of the object code, whose source code is excluded
294 | from the Corresponding Source as a System Library, need not be
295 | included in conveying the object code work.
296 |
297 | A "User Product" is either (1) a "consumer product", which means any
298 | tangible personal property which is normally used for personal, family,
299 | or household purposes, or (2) anything designed or sold for incorporation
300 | into a dwelling. In determining whether a product is a consumer product,
301 | doubtful cases shall be resolved in favor of coverage. For a particular
302 | product received by a particular user, "normally used" refers to a
303 | typical or common use of that class of product, regardless of the status
304 | of the particular user or of the way in which the particular user
305 | actually uses, or expects or is expected to use, the product. A product
306 | is a consumer product regardless of whether the product has substantial
307 | commercial, industrial or non-consumer uses, unless such uses represent
308 | the only significant mode of use of the product.
309 |
310 | "Installation Information" for a User Product means any methods,
311 | procedures, authorization keys, or other information required to install
312 | and execute modified versions of a covered work in that User Product from
313 | a modified version of its Corresponding Source. The information must
314 | suffice to ensure that the continued functioning of the modified object
315 | code is in no case prevented or interfered with solely because
316 | modification has been made.
317 |
318 | If you convey an object code work under this section in, or with, or
319 | specifically for use in, a User Product, and the conveying occurs as
320 | part of a transaction in which the right of possession and use of the
321 | User Product is transferred to the recipient in perpetuity or for a
322 | fixed term (regardless of how the transaction is characterized), the
323 | Corresponding Source conveyed under this section must be accompanied
324 | by the Installation Information. But this requirement does not apply
325 | if neither you nor any third party retains the ability to install
326 | modified object code on the User Product (for example, the work has
327 | been installed in ROM).
328 |
329 | The requirement to provide Installation Information does not include a
330 | requirement to continue to provide support service, warranty, or updates
331 | for a work that has been modified or installed by the recipient, or for
332 | the User Product in which it has been modified or installed. Access to a
333 | network may be denied when the modification itself materially and
334 | adversely affects the operation of the network or violates the rules and
335 | protocols for communication across the network.
336 |
337 | Corresponding Source conveyed, and Installation Information provided,
338 | in accord with this section must be in a format that is publicly
339 | documented (and with an implementation available to the public in
340 | source code form), and must require no special password or key for
341 | unpacking, reading or copying.
342 |
343 | 7. Additional Terms.
344 |
345 | "Additional permissions" are terms that supplement the terms of this
346 | License by making exceptions from one or more of its conditions.
347 | Additional permissions that are applicable to the entire Program shall
348 | be treated as though they were included in this License, to the extent
349 | that they are valid under applicable law. If additional permissions
350 | apply only to part of the Program, that part may be used separately
351 | under those permissions, but the entire Program remains governed by
352 | this License without regard to the additional permissions.
353 |
354 | When you convey a copy of a covered work, you may at your option
355 | remove any additional permissions from that copy, or from any part of
356 | it. (Additional permissions may be written to require their own
357 | removal in certain cases when you modify the work.) You may place
358 | additional permissions on material, added by you to a covered work,
359 | for which you have or can give appropriate copyright permission.
360 |
361 | Notwithstanding any other provision of this License, for material you
362 | add to a covered work, you may (if authorized by the copyright holders of
363 | that material) supplement the terms of this License with terms:
364 |
365 | a) Disclaiming warranty or limiting liability differently from the
366 | terms of sections 15 and 16 of this License; or
367 |
368 | b) Requiring preservation of specified reasonable legal notices or
369 | author attributions in that material or in the Appropriate Legal
370 | Notices displayed by works containing it; or
371 |
372 | c) Prohibiting misrepresentation of the origin of that material, or
373 | requiring that modified versions of such material be marked in
374 | reasonable ways as different from the original version; or
375 |
376 | d) Limiting the use for publicity purposes of names of licensors or
377 | authors of the material; or
378 |
379 | e) Declining to grant rights under trademark law for use of some
380 | trade names, trademarks, or service marks; or
381 |
382 | f) Requiring indemnification of licensors and authors of that
383 | material by anyone who conveys the material (or modified versions of
384 | it) with contractual assumptions of liability to the recipient, for
385 | any liability that these contractual assumptions directly impose on
386 | those licensors and authors.
387 |
388 | All other non-permissive additional terms are considered "further
389 | restrictions" within the meaning of section 10. If the Program as you
390 | received it, or any part of it, contains a notice stating that it is
391 | governed by this License along with a term that is a further
392 | restriction, you may remove that term. If a license document contains
393 | a further restriction but permits relicensing or conveying under this
394 | License, you may add to a covered work material governed by the terms
395 | of that license document, provided that the further restriction does
396 | not survive such relicensing or conveying.
397 |
398 | If you add terms to a covered work in accord with this section, you
399 | must place, in the relevant source files, a statement of the
400 | additional terms that apply to those files, or a notice indicating
401 | where to find the applicable terms.
402 |
403 | Additional terms, permissive or non-permissive, may be stated in the
404 | form of a separately written license, or stated as exceptions;
405 | the above requirements apply either way.
406 |
407 | 8. Termination.
408 |
409 | You may not propagate or modify a covered work except as expressly
410 | provided under this License. Any attempt otherwise to propagate or
411 | modify it is void, and will automatically terminate your rights under
412 | this License (including any patent licenses granted under the third
413 | paragraph of section 11).
414 |
415 | However, if you cease all violation of this License, then your
416 | license from a particular copyright holder is reinstated (a)
417 | provisionally, unless and until the copyright holder explicitly and
418 | finally terminates your license, and (b) permanently, if the copyright
419 | holder fails to notify you of the violation by some reasonable means
420 | prior to 60 days after the cessation.
421 |
422 | Moreover, your license from a particular copyright holder is
423 | reinstated permanently if the copyright holder notifies you of the
424 | violation by some reasonable means, this is the first time you have
425 | received notice of violation of this License (for any work) from that
426 | copyright holder, and you cure the violation prior to 30 days after
427 | your receipt of the notice.
428 |
429 | Termination of your rights under this section does not terminate the
430 | licenses of parties who have received copies or rights from you under
431 | this License. If your rights have been terminated and not permanently
432 | reinstated, you do not qualify to receive new licenses for the same
433 | material under section 10.
434 |
435 | 9. Acceptance Not Required for Having Copies.
436 |
437 | You are not required to accept this License in order to receive or
438 | run a copy of the Program. Ancillary propagation of a covered work
439 | occurring solely as a consequence of using peer-to-peer transmission
440 | to receive a copy likewise does not require acceptance. However,
441 | nothing other than this License grants you permission to propagate or
442 | modify any covered work. These actions infringe copyright if you do
443 | not accept this License. Therefore, by modifying or propagating a
444 | covered work, you indicate your acceptance of this License to do so.
445 |
446 | 10. Automatic Licensing of Downstream Recipients.
447 |
448 | Each time you convey a covered work, the recipient automatically
449 | receives a license from the original licensors, to run, modify and
450 | propagate that work, subject to this License. You are not responsible
451 | for enforcing compliance by third parties with this License.
452 |
453 | An "entity transaction" is a transaction transferring control of an
454 | organization, or substantially all assets of one, or subdividing an
455 | organization, or merging organizations. If propagation of a covered
456 | work results from an entity transaction, each party to that
457 | transaction who receives a copy of the work also receives whatever
458 | licenses to the work the party's predecessor in interest had or could
459 | give under the previous paragraph, plus a right to possession of the
460 | Corresponding Source of the work from the predecessor in interest, if
461 | the predecessor has it or can get it with reasonable efforts.
462 |
463 | You may not impose any further restrictions on the exercise of the
464 | rights granted or affirmed under this License. For example, you may
465 | not impose a license fee, royalty, or other charge for exercise of
466 | rights granted under this License, and you may not initiate litigation
467 | (including a cross-claim or counterclaim in a lawsuit) alleging that
468 | any patent claim is infringed by making, using, selling, offering for
469 | sale, or importing the Program or any portion of it.
470 |
471 | 11. Patents.
472 |
473 | A "contributor" is a copyright holder who authorizes use under this
474 | License of the Program or a work on which the Program is based. The
475 | work thus licensed is called the contributor's "contributor version".
476 |
477 | A contributor's "essential patent claims" are all patent claims
478 | owned or controlled by the contributor, whether already acquired or
479 | hereafter acquired, that would be infringed by some manner, permitted
480 | by this License, of making, using, or selling its contributor version,
481 | but do not include claims that would be infringed only as a
482 | consequence of further modification of the contributor version. For
483 | purposes of this definition, "control" includes the right to grant
484 | patent sublicenses in a manner consistent with the requirements of
485 | this License.
486 |
487 | Each contributor grants you a non-exclusive, worldwide, royalty-free
488 | patent license under the contributor's essential patent claims, to
489 | make, use, sell, offer for sale, import and otherwise run, modify and
490 | propagate the contents of its contributor version.
491 |
492 | In the following three paragraphs, a "patent license" is any express
493 | agreement or commitment, however denominated, not to enforce a patent
494 | (such as an express permission to practice a patent or covenant not to
495 | sue for patent infringement). To "grant" such a patent license to a
496 | party means to make such an agreement or commitment not to enforce a
497 | patent against the party.
498 |
499 | If you convey a covered work, knowingly relying on a patent license,
500 | and the Corresponding Source of the work is not available for anyone
501 | to copy, free of charge and under the terms of this License, through a
502 | publicly available network server or other readily accessible means,
503 | then you must either (1) cause the Corresponding Source to be so
504 | available, or (2) arrange to deprive yourself of the benefit of the
505 | patent license for this particular work, or (3) arrange, in a manner
506 | consistent with the requirements of this License, to extend the patent
507 | license to downstream recipients. "Knowingly relying" means you have
508 | actual knowledge that, but for the patent license, your conveying the
509 | covered work in a country, or your recipient's use of the covered work
510 | in a country, would infringe one or more identifiable patents in that
511 | country that you have reason to believe are valid.
512 |
513 | If, pursuant to or in connection with a single transaction or
514 | arrangement, you convey, or propagate by procuring conveyance of, a
515 | covered work, and grant a patent license to some of the parties
516 | receiving the covered work authorizing them to use, propagate, modify
517 | or convey a specific copy of the covered work, then the patent license
518 | you grant is automatically extended to all recipients of the covered
519 | work and works based on it.
520 |
521 | A patent license is "discriminatory" if it does not include within
522 | the scope of its coverage, prohibits the exercise of, or is
523 | conditioned on the non-exercise of one or more of the rights that are
524 | specifically granted under this License. You may not convey a covered
525 | work if you are a party to an arrangement with a third party that is
526 | in the business of distributing software, under which you make payment
527 | to the third party based on the extent of your activity of conveying
528 | the work, and under which the third party grants, to any of the
529 | parties who would receive the covered work from you, a discriminatory
530 | patent license (a) in connection with copies of the covered work
531 | conveyed by you (or copies made from those copies), or (b) primarily
532 | for and in connection with specific products or compilations that
533 | contain the covered work, unless you entered into that arrangement,
534 | or that patent license was granted, prior to 28 March 2007.
535 |
536 | Nothing in this License shall be construed as excluding or limiting
537 | any implied license or other defenses to infringement that may
538 | otherwise be available to you under applicable patent law.
539 |
540 | 12. No Surrender of Others' Freedom.
541 |
542 | If conditions are imposed on you (whether by court order, agreement or
543 | otherwise) that contradict the conditions of this License, they do not
544 | excuse you from the conditions of this License. If you cannot convey a
545 | covered work so as to satisfy simultaneously your obligations under this
546 | License and any other pertinent obligations, then as a consequence you may
547 | not convey it at all. For example, if you agree to terms that obligate you
548 | to collect a royalty for further conveying from those to whom you convey
549 | the Program, the only way you could satisfy both those terms and this
550 | License would be to refrain entirely from conveying the Program.
551 |
552 | 13. Use with the GNU Affero General Public License.
553 |
554 | Notwithstanding any other provision of this License, you have
555 | permission to link or combine any covered work with a work licensed
556 | under version 3 of the GNU Affero General Public License into a single
557 | combined work, and to convey the resulting work. The terms of this
558 | License will continue to apply to the part which is the covered work,
559 | but the special requirements of the GNU Affero General Public License,
560 | section 13, concerning interaction through a network will apply to the
561 | combination as such.
562 |
563 | 14. Revised Versions of this License.
564 |
565 | The Free Software Foundation may publish revised and/or new versions of
566 | the GNU General Public License from time to time. Such new versions will
567 | be similar in spirit to the present version, but may differ in detail to
568 | address new problems or concerns.
569 |
570 | Each version is given a distinguishing version number. If the
571 | Program specifies that a certain numbered version of the GNU General
572 | Public License "or any later version" applies to it, you have the
573 | option of following the terms and conditions either of that numbered
574 | version or of any later version published by the Free Software
575 | Foundation. If the Program does not specify a version number of the
576 | GNU General Public License, you may choose any version ever published
577 | by the Free Software Foundation.
578 |
579 | If the Program specifies that a proxy can decide which future
580 | versions of the GNU General Public License can be used, that proxy's
581 | public statement of acceptance of a version permanently authorizes you
582 | to choose that version for the Program.
583 |
584 | Later license versions may give you additional or different
585 | permissions. However, no additional obligations are imposed on any
586 | author or copyright holder as a result of your choosing to follow a
587 | later version.
588 |
589 | 15. Disclaimer of Warranty.
590 |
591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
599 |
600 | 16. Limitation of Liability.
601 |
602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
610 | SUCH DAMAGES.
611 |
612 | 17. Interpretation of Sections 15 and 16.
613 |
614 | If the disclaimer of warranty and limitation of liability provided
615 | above cannot be given local legal effect according to their terms,
616 | reviewing courts shall apply local law that most closely approximates
617 | an absolute waiver of all civil liability in connection with the
618 | Program, unless a warranty or assumption of liability accompanies a
619 | copy of the Program in return for a fee.
620 |
621 | END OF TERMS AND CONDITIONS
622 |
623 | How to Apply These Terms to Your New Programs
624 |
625 | If you develop a new program, and you want it to be of the greatest
626 | possible use to the public, the best way to achieve this is to make it
627 | free software which everyone can redistribute and change under these terms.
628 |
629 | To do so, attach the following notices to the program. It is safest
630 | to attach them to the start of each source file to most effectively
631 | state the exclusion of warranty; and each file should have at least
632 | the "copyright" line and a pointer to where the full notice is found.
633 |
634 |
635 | Copyright (C)
636 |
637 | This program is free software: you can redistribute it and/or modify
638 | it under the terms of the GNU General Public License as published by
639 | the Free Software Foundation, either version 3 of the License, or
640 | (at your option) any later version.
641 |
642 | This program is distributed in the hope that it will be useful,
643 | but WITHOUT ANY WARRANTY; without even the implied warranty of
644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
645 | GNU General Public License for more details.
646 |
647 | You should have received a copy of the GNU General Public License
648 | along with this program. If not, see .
649 |
650 | Also add information on how to contact you by electronic and paper mail.
651 |
652 | If the program does terminal interaction, make it output a short
653 | notice like this when it starts in an interactive mode:
654 |
655 | Copyright (C)
656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
657 | This is free software, and you are welcome to redistribute it
658 | under certain conditions; type `show c' for details.
659 |
660 | The hypothetical commands `show w' and `show c' should show the appropriate
661 | parts of the General Public License. Of course, your program's commands
662 | might be different; for a GUI interface, you would use an "about box".
663 |
664 | You should also get your employer (if you work as a programmer) or school,
665 | if any, to sign a "copyright disclaimer" for the program, if necessary.
666 | For more information on this, and how to apply and follow the GNU GPL, see
667 | .
668 |
669 | The GNU General Public License does not permit incorporating your program
670 | into proprietary programs. If your program is a subroutine library, you
671 | may consider it more useful to permit linking proprietary applications with
672 | the library. If this is what you want to do, use the GNU Lesser General
673 | Public License instead of this License. But first, please read
674 | .
675 |
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | English|[简体中文](README_ch.md)
2 |
3 | [](https://colab.research.google.com/drive/1i1hO-_yS6kZdrLden8Mo_hfeHFD-cBt0?usp=sharing)
4 |
5 | # DoFaker: A very simple face swapping tool
6 | Insightface based face swapping tool to replace faces in videos or images. Windows and linux support CPU and GPU. Onnxruntime inference without pytorch.
7 |
8 |
9 | 
10 |
11 |
12 |
13 | 
14 |
15 |
16 |
17 | 
18 | 
19 |
20 |
21 | # Update
22 | - 2023/9/16 update pose transfer model
23 | - 2023/9/14 update face enhance(GFPGAN) and image super resolution(BSRGAN)
24 |
25 | # Tutorial
26 | - [dofaker tutorial in youtube：face swap so easy](https://www.youtube.com/watch?v=qd1-JSpiZao)
27 | - [colab online](https://colab.research.google.com/drive/1i1hO-_yS6kZdrLden8Mo_hfeHFD-cBt0?usp=sharing)
28 |
29 |
30 | # Qiuck Start
31 | install dofaker
32 | ```bash
33 | git clone https://github.com/justld/dofaker.git
34 | cd dofaker
35 | conda create -n dofaker
36 | conda activate dofaker
37 | pip install onnxruntime # onnxruntime-gpu
38 | pip install -e .
39 | ```
40 |
41 | open web ui(The model weights will be downloaded automatically):
42 | ```bash
43 | dofaker
44 | ```
45 |
46 | command line(linux):
47 | ```
48 | bash test.sh
49 | ```
50 |
51 |
52 | # Install from source code
53 | ## 一、Installation
54 | You should install onnxruntime or onnxruntime-gpu manually.
55 |
56 | ### conda install
57 | create virtual environment:
58 | ```bash
59 | git clone https://github.com/justld/dofaker.git
60 | cd dofaker
61 | conda create -n dofaker
62 | conda activate dofaker
63 | pip install -r requirements.txt
64 | pip install onnxruntime # onnxruntime-gpu
65 | pip install -e .
66 | ```
67 |
68 | ### pip install
69 | ```bash
70 | git clone https://github.com/justld/dofaker.git
71 | cd dofaker
72 | pip install -r requirements.txt
73 | pip install onnxruntime # onnxruntime-gpu
74 | pip install -e .
75 | ```
76 |
77 | ## 二、Download Weight
78 | All weights can be downloaded from [release](https://github.com/justld/dofaker/releases). These weight come from links refer to the botton links.
79 |
80 | Unzip the zip file, the dir looks like follow:
81 | ```
82 | |-dofaker
83 | |-docs
84 | |-weights
85 | ----|-models
86 | --------|-buffalo_l
87 | ----------|-1k3d68.onnx
88 | ----------|-2d106det.onnx
89 | ----------|-...
90 | --------|-buffalo_l.zip
91 | --------|-inswapper_128.onnx
92 | --------|-GFPGANv1.3.onnx
93 | --------|-bsrgan_4.onnx
94 | |-run_faceswapper.py
95 | |-web_ui.py
96 | ```
97 |
98 |
99 | ## 三、Usage
100 | You can use dofaker in web_ui or command line.
101 | ### web ui
102 | web gui only support one face swap once, if you want to swap multiple faces, please refer to command usage.
103 | ```bash
104 | python web_ui.py
105 | ```
106 |
107 | ### command
108 | You can swap multiple faces in command.
109 | ```bash
110 | python run_faceswapper.py --source "image or video path to be swapped" --dst_face_paths "dst_face1_path" "dst_face2_path" ... --src_face_paths "src_face1_path" "src_face2_path" ...
111 | ```
112 |
113 | The command follow will replace dst_face1 and dst_face2 detected in input_video.mp4 with src_face1 and src_face2:
114 | ```bash
115 | python run_faceswapper.py --source input_video.mp4 --dst_face_paths dst_face1.jpg dst_face2.jpg --src_face_paths src_face1.jpg src_face2.jpg
116 | ```
117 |
118 | |args|description|
119 | |:---:|:---:|
120 | |source|The image or video to be swapped|
121 | |dst_face_paths|The images includding faces in source to be swapped. If None, replace all faces in source media.|
122 | |src_face_paths|The images includding faces in source to be swapped|
123 |
124 |
125 | # Attention
126 | Do not apply this software to scenarios that violate morality, law, or infringement. The consequences caused by using this software shall be borne by the user themselves.
127 |
128 | # Sponsor
129 | [Thank you for support](https://www.paypal.com/paypalme/justldu)
130 |
131 | # Thanks
132 | - [insightface](https://github.com/deepinsight/insightface)
133 | - [GFPGAN](https://github.com/TencentARC/GFPGAN)
134 | - [GFPGAN-onnxruntime-demo](https://github.com/xuanandsix/GFPGAN-onnxruntime-demo)
135 | - [BSRGAN](https://github.com/cszn/BSRGAN)
136 | - [pose-transfer](https://github.com/prasunroy/pose-transfer)
137 | - [openpose-pytorch](https://github.com/prasunroy/openpose-pytorch)
138 |
--------------------------------------------------------------------------------
/README_ch.md:
--------------------------------------------------------------------------------
1 | [English](README.md)|简体中文
2 |
3 | [](https://aistudio.baidu.com/projectdetail/6759162)
4 |
5 | # DoFaker: 一个简单易用的换脸工具
6 | 基于insightface开发，可以轻松替换视频或图片中的人脸。支持windows和linux系统，CPU和GPU推理。onnxruntime推理，无需pytorch。
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
15 |
16 |
17 |
18 |
19 |
20 |
21 | # 更新
22 | - 2023/9/16 更新动作迁移算法
23 | - 2023/9/14 更新脸部增强算法(GFPGAN)和超分算法(BSRGAN)
24 |
25 | # 教程
26 | - [B站视频使用教程](https://www.bilibili.com/video/BV1b8411i7A8/)
27 | - [AiStudio在线免费体验](https://aistudio.baidu.com/projectdetail/6759162)
28 |
29 |
30 | # 快速开始
31 | 克隆代码，安装dofaker
32 | ```bash
33 | git clone https://github.com/justld/dofaker.git
34 | cd dofaker
35 | conda create -n dofaker
36 | conda activate dofaker
37 | pip install onnxruntime # onnxruntime-gpu
38 | pip install -e .
39 | ```
40 |
41 | 打开web服务（权重自动下载）:
42 | ```bash
43 | dofaker
44 | ```
45 |
46 | 命令行：
47 | ```
48 | bash test.sh
49 | ```
50 |
51 |
52 | ## 源码安装
53 | 手动安装onnxruntime或onnxruntime-gpu.
54 |
55 | ### conda install
56 | 创建conda虚拟环境:
57 | ```bash
58 | git clone https://github.com/justld/dofaker.git
59 | cd dofaker
60 | conda create -n dofaker
61 | conda activate dofaker
62 | pip install -r requirements.txt
63 | pip install onnxruntime # onnxruntime-gpu
64 | ```
65 |
66 | ### pip install
67 | ```bash
68 | git clone https://github.com/justld/dofaker.git
69 | cd dofaker
70 | pip install -r requirements.txt
71 | pip install onnxruntime # onnxruntime-gpu
72 | ```
73 |
74 | ## 二、Download Weight
75 | 所有的权重来自[release](https://github.com/justld/dofaker/releases)，权重来自底部的链接。
76 |
77 | 解压下载好的权重文件，目录结构如下所示:
78 | ```
79 | |-dofaker
80 | |-docs
81 | |-weights
82 | ----|-models
83 | --------|-buffalo_l
84 | ----------|-1k3d68.onnx
85 | ----------|-2d106det.onnx
86 | ----------|-...
87 | --------|-buffalo_l.zip
88 | --------|-inswapper_128.onnx
89 | --------|-GFPGANv1.3.onnx
90 | --------|-bsrgan_4.onnx
91 | |-run_faceswapper.py
92 | |-web_ui.py
93 | ```
94 |
95 |
96 | ## 三、使用
97 | 您可以以web或命令行的方式进行使用
98 | ### web ui
99 | web使用方式只支持单个人脸替换，同时替换多个人脸请使用命令行的方式：
100 | ```bash
101 | python web_ui.py
102 | ```
103 |
104 | ### command
105 | 命令行的使用方法支持一次性多个人脸替换：
106 | ```bash
107 | python run_faceswapper.py --source "image or video path to be swapped" --dst_face_paths "dst_face1_path" "dst_face2_path" ... --src_face_paths "src_face1_path" "src_face2_path" ...
108 | ```
109 |
110 | 以下的命令会使用src_face1和src_face2替换视频input_video.mp4中的dst_face1和dst_face2 ：
111 | ```bash
112 | python run_faceswapper.py --source input_video.mp4 --dst_face_paths dst_face1.jpg dst_face2.jpg --src_face_paths src_face1.jpg src_face2.jpg
113 | ```
114 |
115 | |参数|说明|
116 | |:---:|:---:|
117 | |source|需要替换人脸的图片或视频|
118 | |dst_face_paths|待替换的图片或视频中的目标人脸路径，如果为None，待替换的图片和视频中的所有人脸都被替换为src_face|
119 | |src_face_paths|新的人脸图片路径，用于替换目标图片或视频|
120 |
121 |
122 | # 声明
123 | 禁止将本软件用于违反法律、道德，侵权等场合，本软件仅供学习用途，使用本软件造成的一切后果由使用者承担。
124 |
125 | # 赞助
126 | [您的支持是我们持续开发的动力](https://justld.github.io/)
127 |
128 | # Thanks
129 | - [insightface](https://github.com/deepinsight/insightface)
130 | - [GFPGAN](https://github.com/TencentARC/GFPGAN)
131 | - [GFPGAN-onnxruntime-demo](https://github.com/xuanandsix/GFPGAN-onnxruntime-demo)
132 | - [BSRGAN](https://github.com/cszn/BSRGAN)
133 | - [pose-transfer](https://github.com/prasunroy/pose-transfer)
134 | - [openpose-pytorch](https://github.com/prasunroy/openpose-pytorch)
135 |
--------------------------------------------------------------------------------
/docs/images/andy.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/images/andy.png
--------------------------------------------------------------------------------
/docs/images/multi.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/images/multi.png
--------------------------------------------------------------------------------
/docs/images/source.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/images/source.gif
--------------------------------------------------------------------------------
/docs/images/swapped.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/images/swapped.gif
--------------------------------------------------------------------------------
/docs/images/trump.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/images/trump.jpg
--------------------------------------------------------------------------------
/docs/test/condition.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/condition.jpg
--------------------------------------------------------------------------------
/docs/test/dst1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/dst1.png
--------------------------------------------------------------------------------
/docs/test/dst2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/dst2.png
--------------------------------------------------------------------------------
/docs/test/multi.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/multi.png
--------------------------------------------------------------------------------
/docs/test/taitan.jpeg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/taitan.jpeg
--------------------------------------------------------------------------------
/docs/test/target_pose_reference.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/target_pose_reference.jpg
--------------------------------------------------------------------------------
/docs/test/trump.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/docs/test/trump.jpg
--------------------------------------------------------------------------------
/dofaker/__init__.py:
--------------------------------------------------------------------------------
1 | from .face_det import FaceAnalysis
2 | from .face_swap import InSwapper
3 | from .face_enhance import GFPGAN
4 | from .super_resolution import BSRGAN
5 | from .pose import PoseEstimator
6 | from .face_core import FaceSwapper
7 | from .pose_core import PoseSwapper
8 |
--------------------------------------------------------------------------------
/dofaker/face_core.py:
--------------------------------------------------------------------------------
1 | import os
2 |
3 | import numpy as np
4 | import cv2
5 | from moviepy.editor import VideoFileClip
6 |
7 | from .face_det import FaceAnalysis
8 | from .super_resolution import BSRGAN
9 | from dofaker.face_swap import get_swapper_model
10 | from dofaker.face_enhance import GFPGAN
11 |
12 |
13 | class FaceSwapper:
14 |
15 | def __init__(self,
16 | face_det_model='buffalo_l',
17 | face_swap_model='inswapper',
18 | image_sr_model='bsrgan',
19 | face_enhance_model='gfpgan',
20 | face_det_model_dir='weights/models',
21 | face_swap_model_dir='weights/models',
22 | image_sr_model_dir='weights/models',
23 | face_enhance_model_dir='weights/models',
24 | face_sim_thre=0.5,
25 | log_iters=10,
26 | use_enhancer=True,
27 | use_sr=True,
28 | scale=1):
29 | self.face_sim_thre = face_sim_thre
30 | self.log_iters = log_iters
31 |
32 | self.det_model = FaceAnalysis(name=face_det_model,
33 | root=face_det_model_dir)
34 | self.det_model.prepare(ctx_id=1, det_size=(640, 640))
35 |
36 | self.swapper_model = get_swapper_model(name=face_swap_model,
37 | root=face_swap_model_dir)
38 | if use_enhancer:
39 | self.face_enhance = GFPGAN(name=face_enhance_model,
40 | root=face_enhance_model_dir)
41 | else:
42 | self.face_enhance = None
43 |
44 | if use_sr:
45 | self.sr = BSRGAN(name=image_sr_model,
46 | root=image_sr_model_dir,
47 | scale=scale)
48 | self.scale = scale
49 | else:
50 | self.sr = None
51 | self.scale = scale
52 |
53 | def run(self,
54 | input_path: str,
55 | dst_face_paths,
56 | src_face_paths,
57 | output_dir='output'):
58 | if isinstance(dst_face_paths, str):
59 | dst_face_paths = [dst_face_paths]
60 | if isinstance(src_face_paths, str):
61 | src_face_paths = [src_face_paths]
62 | if input_path.lower().endswith(('jpg', 'jpeg', 'webp', 'png', 'bmp')):
63 | return self.swap_image(input_path, dst_face_paths, src_face_paths,
64 | output_dir)
65 | else:
66 | return self.swap_video(input_path, dst_face_paths, src_face_paths,
67 | output_dir)
68 |
69 | def swap_video(self,
70 | input_video_path,
71 | dst_face_paths,
72 | src_face_paths,
73 | output_dir='output'):
74 | assert os.path.exists(
75 | input_video_path), 'The input video path {} not exist.'
76 | os.makedirs(output_dir, exist_ok=True)
77 | src_faces = self.get_faces(src_face_paths)
78 | if dst_face_paths is not None:
79 | dst_faces = self.get_faces(dst_face_paths)
80 | dst_face_embeddings = self.get_faces_embeddings(dst_faces)
81 | assert len(dst_faces) == len(
82 | src_faces
83 | ), 'The detected faces in source images not equal target image faces.'
84 |
85 | video = cv2.VideoCapture(input_video_path)
86 | fps = video.get(cv2.CAP_PROP_FPS)
87 | total_frames = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
88 | width = int(video.get(cv2.CAP_PROP_FRAME_WIDTH))
89 | height = int(video.get(cv2.CAP_PROP_FRAME_HEIGHT))
90 | frame_size = (width, height)
91 | print('video fps: {}, total_frames: {}, width: {}, height: {}'.format(
92 | fps, total_frames, width, height))
93 |
94 | video_name = os.path.basename(input_video_path).split('.')[0]
95 | four_cc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
96 | temp_video_path = os.path.join(output_dir,
97 | 'temp_{}.mp4'.format(video_name))
98 | save_video_path = os.path.join(output_dir, '{}.mp4'.format(video_name))
99 | output_video = cv2.VideoWriter(
100 | temp_video_path, four_cc, fps,
101 | (int(frame_size[0] * self.scale), int(frame_size[1] * self.scale)))
102 |
103 | i = 0
104 | while video.isOpened():
105 | ret, frame = video.read()
106 | if ret:
107 | if dst_face_paths is not None:
108 | swapped_image = self.swap_faces(frame,
109 | dst_face_embeddings,
110 | src_faces=src_faces)
111 | else:
112 | swapped_image = self.swap_all_faces(frame,
113 | src_faces=src_faces)
114 | i += 1
115 | if i % self.log_iters == 0:
116 | print('processing {}/{}'.format(i, total_frames))
117 | output_video.write(swapped_image)
118 | else:
119 | break
120 |
121 | video.release()
122 | output_video.release()
123 | self.add_audio_to_video(input_video_path, temp_video_path,
124 | save_video_path)
125 | os.remove(temp_video_path)
126 | return save_video_path
127 |
128 | def swap_image(self,
129 | image_path,
130 | dst_face_paths,
131 | src_face_paths,
132 | output_dir='output'):
133 | os.makedirs(output_dir, exist_ok=True)
134 | src_faces = self.get_faces(src_face_paths)
135 | if dst_face_paths is not None:
136 | dst_faces = self.get_faces(dst_face_paths)
137 | dst_face_embeddings = self.get_faces_embeddings(dst_faces)
138 | assert len(dst_faces) == len(
139 | src_faces
140 | ), 'The detected faces in source images not equal target image faces.'
141 |
142 | image = cv2.imread(image_path)
143 | if dst_face_paths is not None:
144 | swapped_image = self.swap_faces(image,
145 | dst_face_embeddings,
146 | src_faces=src_faces)
147 | else:
148 | swapped_image = self.swap_all_faces(image, src_faces=src_faces)
149 | base_name = os.path.basename(image_path)
150 | save_path = os.path.join(output_dir, base_name)
151 | cv2.imwrite(save_path, swapped_image)
152 | return save_path
153 |
154 | def add_audio_to_video(self, src_video_path, target_video_path,
155 | save_video_path):
156 | audio = VideoFileClip(src_video_path).audio
157 | target_video = VideoFileClip(target_video_path)
158 | target_video = target_video.set_audio(audio)
159 | target_video.write_videofile(save_video_path)
160 | return target_video_path
161 |
162 | def get_faces(self, image_paths):
163 | if isinstance(image_paths, str):
164 | image_paths = [image_paths]
165 | faces = []
166 | for image_path in image_paths:
167 | image = cv2.imread(image_path)
168 | assert image is not None, "the source image is None, please check your image {} format.".format(
169 | image_path)
170 | img_faces = self.det_model.get(image, max_num=1)
171 | assert len(
172 | img_faces
173 | ) == 1, 'The detected face in image {} must be 1, but got {}, please ensure your image including one face.'.format(
174 | image_path, len(img_faces))
175 | faces += img_faces
176 | return faces
177 |
178 | def swap_faces(self, image, dst_face_embeddings: np.ndarray,
179 | src_faces: list) -> np.ndarray:
180 | res = image.copy()
181 | image_faces = self.det_model.get(image)
182 | if len(image_faces) == 0:
183 | return res
184 | image_face_embeddings = self.get_faces_embeddings(image_faces)
185 | sim = np.dot(dst_face_embeddings, image_face_embeddings.T)
186 |
187 | for i in range(dst_face_embeddings.shape[0]):
188 | index = np.where(sim[i] > self.face_sim_thre)[0].tolist()
189 | for idx in index:
190 | res = self.swapper_model.get(res,
191 | image_faces[idx],
192 | src_faces[i],
193 | paste_back=True)
194 | if self.face_enhance is not None:
195 | res = self.face_enhance.get(res,
196 | image_faces[idx],
197 | paste_back=True)
198 |
199 | if self.sr is not None:
200 | res = self.sr.get(res, image_format='bgr')
201 | return res
202 |
203 | def swap_all_faces(self, image, src_faces: list) -> np.ndarray:
204 | assert len(
205 | src_faces
206 | ) == 1, 'If replace all faces in source, the number of src face should be 1, but got {}.'.format(
207 | len(src_faces))
208 | res = image.copy()
209 | image_faces = self.det_model.get(image)
210 | if len(image_faces) == 0:
211 | return res
212 | for image_face in image_faces:
213 | res = self.swapper_model.get(res,
214 | image_face,
215 | src_faces[0],
216 | paste_back=True)
217 | if self.face_enhance is not None:
218 | res = self.face_enhance.get(res, image_face, paste_back=True)
219 | if self.sr is not None:
220 | res = self.sr.get(res, image_format='bgr')
221 | return res
222 |
223 | def get_faces_embeddings(self, faces):
224 | feats = []
225 | for face in faces:
226 | feats.append(face.normed_embedding)
227 | if len(feats) == 1:
228 | feats = np.array(feats, dtype=np.float32).reshape(1, -1)
229 | else:
230 | feats = np.array(feats, dtype=np.float32)
231 | return feats
232 |
--------------------------------------------------------------------------------
/dofaker/face_det/__init__.py:
--------------------------------------------------------------------------------
1 | from .face_analysis import FaceAnalysis
2 |
--------------------------------------------------------------------------------
/dofaker/face_det/face_analysis.py:
--------------------------------------------------------------------------------
1 | '''
2 | The following code references:: https://github.com/deepinsight/insightface
3 | '''
4 |
5 | import glob
6 | import os.path as osp
7 |
8 | import onnxruntime
9 |
10 | from insightface import model_zoo
11 | from insightface.utils import ensure_available
12 | from insightface.app.common import Face
13 |
14 | from dofaker.utils import download_file, get_model_url
15 |
16 | __all__ = ['FaceAnalysis']
17 |
18 |
19 | class FaceAnalysis:
20 |
21 | def __init__(self,
22 | name='buffalo_l',
23 | root='weights',
24 | allowed_modules=None,
25 | **kwargs):
26 | self.model_dir, _ = download_file(get_model_url(name),
27 | save_dir=root,
28 | overwrite=False)
29 | onnxruntime.set_default_logger_severity(3)
30 |
31 | self.models = {}
32 | print(self.model_dir)
33 | onnx_files = glob.glob(osp.join(self.model_dir, '*.onnx'))
34 | onnx_files = sorted(onnx_files)
35 | for onnx_file in onnx_files:
36 | model = model_zoo.get_model(onnx_file, **kwargs)
37 | if model is None:
38 | print('model not recognized:', onnx_file)
39 | elif allowed_modules is not None and model.taskname not in allowed_modules:
40 | print('model ignore:', onnx_file, model.taskname)
41 | del model
42 | elif model.taskname not in self.models and (allowed_modules is None
43 | or model.taskname
44 | in allowed_modules):
45 | print('find model:', onnx_file, model.taskname,
46 | model.input_shape, model.input_mean, model.input_std)
47 | self.models[model.taskname] = model
48 | else:
49 | print('duplicated model task type, ignore:', onnx_file,
50 | model.taskname)
51 | del model
52 | assert 'detection' in self.models
53 | self.det_model = self.models['detection']
54 |
55 | def prepare(self, ctx_id, det_thresh=0.5, det_size=(640, 640)):
56 | self.det_thresh = det_thresh
57 | assert det_size is not None, "det_size can't be None."
58 | self.det_size = det_size
59 | for taskname, model in self.models.items():
60 | if taskname == 'detection':
61 | model.prepare(ctx_id,
62 | input_size=det_size,
63 | det_thresh=det_thresh)
64 | else:
65 | model.prepare(ctx_id)
66 |
67 | def get(self, img, max_num=0):
68 | bboxes, kpss = self.det_model.detect(img,
69 | max_num=max_num,
70 | metric='default')
71 | if bboxes.shape[0] == 0:
72 | return []
73 | ret = []
74 | for i in range(bboxes.shape[0]):
75 | bbox = bboxes[i, 0:4]
76 | det_score = bboxes[i, 4]
77 | kps = None
78 | if kpss is not None:
79 | kps = kpss[i]
80 | face = Face(bbox=bbox, kps=kps, det_score=det_score)
81 | for taskname, model in self.models.items():
82 | if taskname == 'detection':
83 | continue
84 | model.get(img, face)
85 | ret.append(face)
86 | return ret
87 |
88 | def draw_on(self, img, faces):
89 | import cv2
90 | dimg = img.copy()
91 | for i in range(len(faces)):
92 | face = faces[i]
93 | box = face.bbox.astype('int')
94 | color = (0, 0, 255)
95 | cv2.rectangle(dimg, (box[0], box[1]), (box[2], box[3]), color, 2)
96 | if face.kps is not None:
97 | kps = face.kps.astype('int')
98 | for l in range(kps.shape[0]):
99 | color = (0, 0, 255)
100 | if l == 0 or l == 3:
101 | color = (0, 255, 0)
102 | cv2.circle(dimg, (kps[l][0], kps[l][1]), 1, color, 2)
103 | if face.gender is not None and face.age is not None:
104 | cv2.putText(dimg, '%s,%d' % (face.sex, face.age),
105 | (box[0] - 1, box[1] - 4), cv2.FONT_HERSHEY_COMPLEX,
106 | 0.7, (0, 255, 0), 1)
107 | return dimg
108 |
--------------------------------------------------------------------------------
/dofaker/face_enhance/__init__.py:
--------------------------------------------------------------------------------
1 | from .gfpgan import GFPGAN
2 |
--------------------------------------------------------------------------------
/dofaker/face_enhance/gfpgan.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 |
3 | import cv2
4 |
5 | from insightface.utils import face_align
6 | from insightface import model_zoo
7 | from dofaker.utils import download_file, get_model_url
8 |
9 |
10 | class GFPGAN:
11 |
12 | def __init__(self, name='gfpgan', root='weights/models') -> None:
13 | _, model_file = download_file(get_model_url(name),
14 | save_dir=root,
15 | overwrite=False)
16 | providers = model_zoo.model_zoo.get_default_providers()
17 | self.session = model_zoo.model_zoo.PickableInferenceSession(
18 | model_file, providers=providers)
19 |
20 | self.input_mean = 127.5
21 | self.input_std = 127.5
22 | inputs = self.session.get_inputs()
23 | self.input_names = []
24 | for inp in inputs:
25 | self.input_names.append(inp.name)
26 | outputs = self.session.get_outputs()
27 | output_names = []
28 | for out in outputs:
29 | output_names.append(out.name)
30 | self.output_names = output_names
31 | assert len(
32 | self.output_names
33 | ) == 1, "The output number of GFPGAN model should be 1, but got {}, please check your model.".format(
34 | len(self.output_names))
35 | output_shape = outputs[0].shape
36 | input_cfg = inputs[0]
37 | input_shape = input_cfg.shape
38 | self.input_shape = input_shape
39 | print('face_enhance-shape:', self.input_shape)
40 | self.input_size = tuple(input_shape[2:4][::-1])
41 |
42 | def forward(self, image, image_format='bgr'):
43 | if isinstance(image, str):
44 | image = cv2.imread(image, 1)
45 | elif isinstance(image, np.ndarray):
46 | if image_format == 'bgr':
47 | image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
48 | elif image_format == 'rgb':
49 | pass
50 | else:
51 | raise UserWarning(
52 | "gfpgan not support image format {}".format(image_format))
53 | else:
54 | raise UserWarning(
55 | "gfpgan input must be str or np.ndarray, but got {}.".format(
56 | type(image)))
57 | img = (image - self.input_mean) / self.input_std
58 | pred = self.session.run(self.output_names,
59 | {self.input_names[0]: img})[0]
60 | return pred
61 |
62 | def _get(self, img, image_format='bgr'):
63 | if image_format.lower() == 'bgr':
64 | img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
65 | elif image_format.lower() == 'rgb':
66 | pass
67 | else:
68 | raise UserWarning(
69 | "gfpgan not support image format {}".format(image_format))
70 | h, w, c = img.shape
71 | img = cv2.resize(img, (self.input_shape[-1], self.input_shape[-2]))
72 | blob = cv2.dnn.blobFromImage(
73 | img,
74 | 1.0 / self.input_std,
75 | self.input_size,
76 | (self.input_mean, self.input_mean, self.input_mean),
77 | swapRB=False)
78 | pred = self.session.run(self.output_names,
79 | {self.input_names[0]: blob})[0]
80 | image_aug = pred.transpose((0, 2, 3, 1))[0]
81 | rgb_aug = np.clip(self.input_std * image_aug + self.input_mean, 0,
82 | 255).astype(np.uint8)
83 | rgb_aug = cv2.resize(rgb_aug, (w, h))
84 | bgr_image = rgb_aug[:, :, ::-1]
85 | return bgr_image
86 |
87 | def get(self, img, target_face, paste_back=True, image_format='bgr'):
88 | aimg, M = face_align.norm_crop2(img, target_face.kps,
89 | self.input_size[0])
90 | bgr_fake = self._get(aimg, image_format='bgr')
91 | if not paste_back:
92 | return bgr_fake, M
93 | else:
94 | target_img = img
95 | fake_diff = bgr_fake.astype(np.float32) - aimg.astype(np.float32)
96 | fake_diff = np.abs(fake_diff).mean(axis=2)
97 | fake_diff[:2, :] = 0
98 | fake_diff[-2:, :] = 0
99 | fake_diff[:, :2] = 0
100 | fake_diff[:, -2:] = 0
101 | IM = cv2.invertAffineTransform(M)
102 | img_white = np.full((aimg.shape[0], aimg.shape[1]),
103 | 255,
104 | dtype=np.float32)
105 | bgr_fake = cv2.warpAffine(
106 | bgr_fake,
107 | IM, (target_img.shape[1], target_img.shape[0]),
108 | borderValue=0.0)
109 | img_white = cv2.warpAffine(
110 | img_white,
111 | IM, (target_img.shape[1], target_img.shape[0]),
112 | borderValue=0.0)
113 | fake_diff = cv2.warpAffine(
114 | fake_diff,
115 | IM, (target_img.shape[1], target_img.shape[0]),
116 | borderValue=0.0)
117 | img_white[img_white > 20] = 255
118 | fthresh = 10
119 | fake_diff[fake_diff < fthresh] = 0
120 | fake_diff[fake_diff >= fthresh] = 255
121 | img_mask = img_white
122 | mask_h_inds, mask_w_inds = np.where(img_mask == 255)
123 | mask_h = np.max(mask_h_inds) - np.min(mask_h_inds)
124 | mask_w = np.max(mask_w_inds) - np.min(mask_w_inds)
125 | mask_size = int(np.sqrt(mask_h * mask_w))
126 | k = max(mask_size // 10, 10)
127 | #k = max(mask_size//20, 6)
128 | #k = 6
129 | kernel = np.ones((k, k), np.uint8)
130 | img_mask = cv2.erode(img_mask, kernel, iterations=1)
131 | kernel = np.ones((2, 2), np.uint8)
132 | fake_diff = cv2.dilate(fake_diff, kernel, iterations=1)
133 | k = max(mask_size // 20, 5)
134 | kernel_size = (k, k)
135 | blur_size = tuple(2 * i + 1 for i in kernel_size)
136 | img_mask = cv2.GaussianBlur(img_mask, blur_size, 0)
137 | k = 5
138 | kernel_size = (k, k)
139 | blur_size = tuple(2 * i + 1 for i in kernel_size)
140 | fake_diff = cv2.GaussianBlur(fake_diff, blur_size, 0)
141 | img_mask /= 255
142 | fake_diff /= 255
143 | img_mask = np.reshape(img_mask,
144 | [img_mask.shape[0], img_mask.shape[1], 1])
145 | fake_merged = img_mask * bgr_fake + (
146 | 1 - img_mask) * target_img.astype(np.float32)
147 | fake_merged = fake_merged.astype(np.uint8)
148 | return fake_merged
149 |
--------------------------------------------------------------------------------
/dofaker/face_swap/__init__.py:
--------------------------------------------------------------------------------
1 | from .inswapper import InSwapper
2 |
3 |
4 | def get_swapper_model(name='', root=None, **kwargs):
5 | if name.lower() == 'inswapper':
6 | return InSwapper(name=name, root=root, **kwargs)
7 | else:
8 | raise UserWarning('The swapper model {} not support.'.format(name))
9 |
--------------------------------------------------------------------------------
/dofaker/face_swap/base_swapper.py:
--------------------------------------------------------------------------------
1 | class BaseSwapper:
2 |
3 | def forward(self, img, latent, *args, **kwargs):
4 | raise NotImplementedError
5 |
6 | def get(self,
7 | img,
8 | target_face,
9 | source_face,
10 | paste_back=True,
11 | *args,
12 | **kwargs):
13 | raise NotImplementedError
14 |
--------------------------------------------------------------------------------
/dofaker/face_swap/inswapper.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 | import cv2
3 | import onnx
4 | from onnx import numpy_helper
5 |
6 | from insightface import model_zoo
7 | from insightface.utils import face_align
8 | from .base_swapper import BaseSwapper
9 |
10 | from dofaker.utils import download_file, get_model_url
11 |
12 |
13 | class InSwapper(BaseSwapper):
14 |
15 | def __init__(self, name='inswapper', root='weights/models'):
16 | _, model_file = download_file(get_model_url(name),
17 | save_dir=root,
18 | overwrite=False)
19 | providers = model_zoo.model_zoo.get_default_providers()
20 | self.session = model_zoo.model_zoo.PickableInferenceSession(
21 | model_file, providers=providers)
22 |
23 | model = onnx.load(model_file)
24 | graph = model.graph
25 | self.emap = numpy_helper.to_array(graph.initializer[-1])
26 | self.input_mean = 0.0
27 | self.input_std = 255.0
28 |
29 | inputs = self.session.get_inputs()
30 | self.input_names = []
31 | for inp in inputs:
32 | self.input_names.append(inp.name)
33 | outputs = self.session.get_outputs()
34 | output_names = []
35 | for out in outputs:
36 | output_names.append(out.name)
37 | self.output_names = output_names
38 | assert len(
39 | self.output_names
40 | ) == 1, "The output number of inswapper model should be 1, but got {}, please check your model.".format(
41 | len(self.output_names))
42 | output_shape = outputs[0].shape
43 | input_cfg = inputs[0]
44 | input_shape = input_cfg.shape
45 | self.input_shape = input_shape
46 | print('inswapper-shape:', self.input_shape)
47 | self.input_size = tuple(input_shape[2:4][::-1])
48 |
49 | def forward(self, img, latent):
50 | img = (img - self.input_mean) / self.input_std
51 | pred = self.session.run(self.output_names, {
52 | self.input_names[0]: img,
53 | self.input_names[1]: latent
54 | })[0]
55 | return pred
56 |
57 | def get(self, img, target_face, source_face, paste_back=True):
58 | aimg, M = face_align.norm_crop2(img, target_face.kps,
59 | self.input_size[0])
60 | blob = cv2.dnn.blobFromImage(
61 | aimg,
62 | 1.0 / self.input_std,
63 | self.input_size,
64 | (self.input_mean, self.input_mean, self.input_mean),
65 | swapRB=True)
66 | latent = source_face.normed_embedding.reshape((1, -1))
67 | latent = np.dot(latent, self.emap)
68 | latent /= np.linalg.norm(latent)
69 | pred = self.session.run(self.output_names, {
70 | self.input_names[0]: blob,
71 | self.input_names[1]: latent
72 | })[0]
73 | img_fake = pred.transpose((0, 2, 3, 1))[0]
74 | bgr_fake = np.clip(255 * img_fake, 0, 255).astype(np.uint8)[:, :, ::-1]
75 | if not paste_back:
76 | return bgr_fake, M
77 | else:
78 | target_img = img
79 | fake_diff = bgr_fake.astype(np.float32) - aimg.astype(np.float32)
80 | fake_diff = np.abs(fake_diff).mean(axis=2)
81 | fake_diff[:2, :] = 0
82 | fake_diff[-2:, :] = 0
83 | fake_diff[:, :2] = 0
84 | fake_diff[:, -2:] = 0
85 | IM = cv2.invertAffineTransform(M)
86 | img_white = np.full((aimg.shape[0], aimg.shape[1]),
87 | 255,
88 | dtype=np.float32)
89 | bgr_fake = cv2.warpAffine(
90 | bgr_fake,
91 | IM, (target_img.shape[1], target_img.shape[0]),
92 | borderValue=0.0)
93 | img_white = cv2.warpAffine(
94 | img_white,
95 | IM, (target_img.shape[1], target_img.shape[0]),
96 | borderValue=0.0)
97 | fake_diff = cv2.warpAffine(
98 | fake_diff,
99 | IM, (target_img.shape[1], target_img.shape[0]),
100 | borderValue=0.0)
101 | img_white[img_white > 20] = 255
102 | fthresh = 10
103 | fake_diff[fake_diff < fthresh] = 0
104 | fake_diff[fake_diff >= fthresh] = 255
105 | img_mask = img_white
106 | mask_h_inds, mask_w_inds = np.where(img_mask == 255)
107 | mask_h = np.max(mask_h_inds) - np.min(mask_h_inds)
108 | mask_w = np.max(mask_w_inds) - np.min(mask_w_inds)
109 | mask_size = int(np.sqrt(mask_h * mask_w))
110 | k = max(mask_size // 10, 10)
111 | #k = max(mask_size//20, 6)
112 | #k = 6
113 | kernel = np.ones((k, k), np.uint8)
114 | img_mask = cv2.erode(img_mask, kernel, iterations=1)
115 | kernel = np.ones((2, 2), np.uint8)
116 | fake_diff = cv2.dilate(fake_diff, kernel, iterations=1)
117 | k = max(mask_size // 20, 5)
118 | #k = 3
119 | #k = 3
120 | kernel_size = (k, k)
121 | blur_size = tuple(2 * i + 1 for i in kernel_size)
122 | img_mask = cv2.GaussianBlur(img_mask, blur_size, 0)
123 | k = 5
124 | kernel_size = (k, k)
125 | blur_size = tuple(2 * i + 1 for i in kernel_size)
126 | fake_diff = cv2.GaussianBlur(fake_diff, blur_size, 0)
127 | img_mask /= 255
128 | fake_diff /= 255
129 | #img_mask = fake_diff
130 | img_mask = np.reshape(img_mask,
131 | [img_mask.shape[0], img_mask.shape[1], 1])
132 | fake_merged = img_mask * bgr_fake + (
133 | 1 - img_mask) * target_img.astype(np.float32)
134 | fake_merged = fake_merged.astype(np.uint8)
135 | return fake_merged
136 |
--------------------------------------------------------------------------------
/dofaker/pose/__init__.py:
--------------------------------------------------------------------------------
1 | from .pose_estimator import PoseEstimator
2 | from .pose_transfer import PoseTransfer
3 |
--------------------------------------------------------------------------------
/dofaker/pose/pose_estimator.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 |
3 | import cv2
4 | from scipy.ndimage.filters import gaussian_filter
5 |
6 | from .pose_utils import _get_keypoints, _pad_image
7 | from insightface import model_zoo
8 | from dofaker.utils import download_file, get_model_url
9 |
10 |
11 | class PoseEstimator:
12 |
13 | def __init__(self, name='openpose_body', root='weights/models'):
14 | _, model_file = download_file(get_model_url(name),
15 | save_dir=root,
16 | overwrite=False)
17 | providers = model_zoo.model_zoo.get_default_providers()
18 | self.session = model_zoo.model_zoo.PickableInferenceSession(
19 | model_file, providers=providers)
20 |
21 | self.input_mean = 127.5
22 | self.input_std = 255.0
23 | inputs = self.session.get_inputs()
24 | self.input_names = []
25 | for inp in inputs:
26 | self.input_names.append(inp.name)
27 | outputs = self.session.get_outputs()
28 | output_names = []
29 | for out in outputs:
30 | output_names.append(out.name)
31 | self.output_names = output_names
32 | assert len(
33 | self.output_names
34 | ) == 2, "The output number of PoseEstimator model should be 2, but got {}, please check your model.".format(
35 | len(self.output_names))
36 | output_shape = outputs[0].shape
37 | input_cfg = inputs[0]
38 | input_shape = input_cfg.shape
39 | self.input_shape = input_shape
40 | print('pose estimator shape:', self.input_shape)
41 |
42 | def forward(self, image, image_format='rgb'):
43 | if isinstance(image, str):
44 | image = cv2.imread(image, 1)
45 | image_format = 'bgr'
46 | elif isinstance(image, np.ndarray):
47 | if image_format == 'bgr':
48 | pass
49 | elif image_format == 'rgb':
50 | image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
51 | image_format = 'bgr'
52 | else:
53 | raise UserWarning(
54 | "PoseEstimator not support image format {}".format(
55 | image_format))
56 | else:
57 | raise UserWarning(
58 | "PoseEstimator input must be str or np.ndarray, but got {}.".
59 | format(type(image)))
60 |
61 | scales = [0.5]
62 | stride = 8
63 | bboxsize = 368
64 | padvalue = 128
65 | thresh_1 = 0.1
66 | thresh_2 = 0.05
67 |
68 | multipliers = [scale * bboxsize / image.shape[0] for scale in scales]
69 | heatmap_avg = np.zeros((image.shape[0], image.shape[1], 19))
70 | paf_avg = np.zeros((image.shape[0], image.shape[1], 38))
71 |
72 | for scale in multipliers:
73 | image_scaled = cv2.resize(image, (0, 0),
74 | fx=scale,
75 | fy=scale,
76 | interpolation=cv2.INTER_CUBIC)
77 | image_padded, pads = _pad_image(image_scaled, stride, padvalue)
78 |
79 | image_tensor = np.expand_dims(np.transpose(image_padded, (2, 0, 1)),
80 | 0)
81 | blob = (np.float32(image_tensor) - self.input_mean) / self.input_std
82 |
83 | pred = self.session.run(self.output_names,
84 | {self.input_names[0]: blob})
85 | Mconv7_stage6_L1, Mconv7_stage6_L2 = pred[0], pred[1]
86 |
87 | heatmap = np.transpose(np.squeeze(Mconv7_stage6_L2), (1, 2, 0))
88 | heatmap = cv2.resize(heatmap, (0, 0),
89 | fx=stride,
90 | fy=stride,
91 | interpolation=cv2.INTER_CUBIC)
92 | heatmap = heatmap[:image_padded.shape[0] -
93 | pads[3], :image_padded.shape[1] - pads[2], :]
94 | heatmap = cv2.resize(heatmap, (image.shape[1], image.shape[0]),
95 | interpolation=cv2.INTER_CUBIC)
96 |
97 | paf = np.transpose(np.squeeze(Mconv7_stage6_L1), (1, 2, 0))
98 | paf = cv2.resize(paf, (0, 0),
99 | fx=stride,
100 | fy=stride,
101 | interpolation=cv2.INTER_CUBIC)
102 | paf = paf[:image_padded.shape[0] - pads[3], :image_padded.shape[1] -
103 | pads[2], :]
104 | paf = cv2.resize(paf, (image.shape[1], image.shape[0]),
105 | interpolation=cv2.INTER_CUBIC)
106 |
107 | heatmap_avg += (heatmap / len(multipliers))
108 | paf_avg += (paf / len(multipliers))
109 |
110 | all_peaks = []
111 | num_peaks = 0
112 |
113 | for part in range(18):
114 | map_orig = heatmap_avg[:, :, part]
115 | map_filt = gaussian_filter(map_orig, sigma=3)
116 |
117 | map_L = np.zeros_like(map_filt)
118 | map_T = np.zeros_like(map_filt)
119 | map_R = np.zeros_like(map_filt)
120 | map_B = np.zeros_like(map_filt)
121 | map_L[1:, :] = map_filt[:-1, :]
122 | map_T[:, 1:] = map_filt[:, :-1]
123 | map_R[:-1, :] = map_filt[1:, :]
124 | map_B[:, :-1] = map_filt[:, 1:]
125 |
126 | peaks_binary = np.logical_and.reduce(
127 | (map_filt >= map_L, map_filt >= map_T, map_filt
128 | >= map_R, map_filt >= map_B, map_filt > thresh_1))
129 | peaks = list(
130 | zip(np.nonzero(peaks_binary)[1],
131 | np.nonzero(peaks_binary)[0]))
132 | peaks_ids = range(num_peaks, num_peaks + len(peaks))
133 | peaks_with_scores = [
134 | peak + (map_orig[peak[1], peak[0]], ) for peak in peaks
135 | ]
136 | peaks_with_scores_and_ids = [peaks_with_scores[i] + (peaks_ids[i],) \
137 | for i in range(len(peaks_ids))]
138 | all_peaks.append(peaks_with_scores_and_ids)
139 | num_peaks += len(peaks)
140 |
141 | map_idx = [[31, 32], [39, 40], [33, 34], [35, 36], [41, 42], [43, 44],
142 | [19, 20], [21, 22], [23, 24], [25, 26], [27, 28], [29, 30],
143 | [47, 48], [49, 50], [53, 54], [51, 52], [55, 56], [37, 38],
144 | [45, 46]]
145 | limbseq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9],
146 | [9, 10], [10, 11], [2, 12], [12, 13], [13, 14], [2, 1],
147 | [1, 15], [15, 17], [1, 16], [16, 18], [3, 17], [6, 18]]
148 |
149 | all_connections = []
150 | spl_k = []
151 | mid_n = 10
152 |
153 | for k in range(len(map_idx)):
154 | score_mid = paf_avg[:, :, [x - 19 for x in map_idx[k]]]
155 | candidate_A = all_peaks[limbseq[k][0] - 1]
156 | candidate_B = all_peaks[limbseq[k][1] - 1]
157 | n_A = len(candidate_A)
158 | n_B = len(candidate_B)
159 | index_A, index_B = limbseq[k]
160 | if n_A != 0 and n_B != 0:
161 | connection_candidates = []
162 | for i in range(n_A):
163 | for j in range(n_B):
164 | v = np.subtract(candidate_B[j][:2], candidate_A[i][:2])
165 | n = np.sqrt(v[0] * v[0] + v[1] * v[1])
166 | v = np.divide(v, n)
167 |
168 | ab = list(
169 | zip(
170 | np.linspace(candidate_A[i][0],
171 | candidate_B[j][0],
172 | num=mid_n),
173 | np.linspace(candidate_A[i][1],
174 | candidate_B[j][1],
175 | num=mid_n)))
176 | vx = np.array([
177 | score_mid[int(round(ab[x][1])),
178 | int(round(ab[x][0])), 0]
179 | for x in range(len(ab))
180 | ])
181 | vy = np.array([
182 | score_mid[int(round(ab[x][1])),
183 | int(round(ab[x][0])), 1]
184 | for x in range(len(ab))
185 | ])
186 | score_midpoints = np.multiply(vx, v[0]) + np.multiply(
187 | vy, v[1])
188 | score_with_dist_prior = sum(
189 | score_midpoints) / len(score_midpoints) + min(
190 | 0.5 * image.shape[0] / n - 1, 0)
191 | criterion_1 = len(
192 | np.nonzero(score_midpoints > thresh_2)
193 | [0]) > 0.8 * len(score_midpoints)
194 | criterion_2 = score_with_dist_prior > 0
195 | if criterion_1 and criterion_2:
196 | connection_candidate = [
197 | i, j, score_with_dist_prior,
198 | score_with_dist_prior + candidate_A[i][2] +
199 | candidate_B[j][2]
200 | ]
201 | connection_candidates.append(connection_candidate)
202 | connection_candidates = sorted(connection_candidates,
203 | key=lambda x: x[2],
204 | reverse=True)
205 | connection = np.zeros((0, 5))
206 | for candidate in connection_candidates:
207 | i, j, s = candidate[0:3]
208 | if i not in connection[:, 3] and j not in connection[:, 4]:
209 | connection = np.vstack([
210 | connection,
211 | [candidate_A[i][3], candidate_B[j][3], s, i, j]
212 | ])
213 | if len(connection) >= min(n_A, n_B):
214 | break
215 | all_connections.append(connection)
216 | else:
217 | spl_k.append(k)
218 | all_connections.append([])
219 |
220 | candidate = np.array(
221 | [item for sublist in all_peaks for item in sublist])
222 | subset = np.ones((0, 20)) * -1
223 |
224 | for k in range(len(map_idx)):
225 | if k not in spl_k:
226 | part_As = all_connections[k][:, 0]
227 | part_Bs = all_connections[k][:, 1]
228 | index_A, index_B = np.array(limbseq[k]) - 1
229 | for i in range(len(all_connections[k])):
230 | found = 0
231 | subset_idx = [-1, -1]
232 | for j in range(len(subset)):
233 | if subset[j][index_A] == part_As[i] or subset[j][
234 | index_B] == part_Bs[i]:
235 | subset_idx[found] = j
236 | found += 1
237 | if found == 1:
238 | j = subset_idx[0]
239 | if subset[j][index_B] != part_Bs[i]:
240 | subset[j][index_B] = part_Bs[i]
241 | subset[j][-1] += 1
242 | subset[j][-2] += candidate[
243 | part_Bs[i].astype(int),
244 | 2] + all_connections[k][i][2]
245 | elif found == 2:
246 | j1, j2 = subset_idx
247 | membership = ((subset[j1] >= 0).astype(int) +
248 | (subset[j2] >= 0).astype(int))[:-2]
249 | if len(np.nonzero(membership == 2)[0]) == 0:
250 | subset[j1][:-2] += (subset[j2][:-2] + 1)
251 | subset[j1][-2:] += subset[j2][-2:]
252 | subset[j1][-2] += all_connections[k][i][2]
253 | subset = np.delete(subset, j2, 0)
254 | else:
255 | subset[j1][index_B] = part_Bs[i]
256 | subset[j1][-1] += 1
257 | subset[j1][-2] += candidate[
258 | part_Bs[i].astype(int),
259 | 2] + all_connections[k][i][2]
260 | elif not found and k < 17:
261 | row = np.ones(20) * -1
262 | row[index_A] = part_As[i]
263 | row[index_B] = part_Bs[i]
264 | row[-1] = 2
265 | row[-2] = sum(
266 | candidate[all_connections[k][i, :2].astype(int),
267 | 2]) + all_connections[k][i][2]
268 | subset = np.vstack([subset, row])
269 |
270 | del_idx = []
271 |
272 | for i in range(len(subset)):
273 | if subset[i][-1] < 4 or subset[i][-2] / subset[i][-1] < 0.4:
274 | del_idx.append(i)
275 | subset = np.delete(subset, del_idx, axis=0)
276 |
277 | return _get_keypoints(candidate, subset)
278 |
279 | def get(self, image, image_format='rgb'):
280 | return self.forward(image, image_format)
281 |
--------------------------------------------------------------------------------
/dofaker/pose/pose_transfer.py:
--------------------------------------------------------------------------------
1 | import cv2
2 | import numpy as np
3 | from scipy.ndimage.filters import gaussian_filter
4 |
5 | from .pose_utils import _get_keypoints, _pad_image
6 | from insightface import model_zoo
7 | from dofaker.utils import download_file, get_model_url
8 | from dofaker.transforms import center_crop, pad
9 |
10 |
11 | class PoseTransfer:
12 |
13 | def __init__(self,
14 | name='pose_transfer',
15 | root='weights/models',
16 | pose_estimator=None):
17 | assert pose_estimator is not None, "The pose_estimator of PoseTransfer shouldn't be None"
18 | self.pose_estimator = pose_estimator
19 | _, model_file = download_file(get_model_url(name),
20 | save_dir=root,
21 | overwrite=False)
22 | providers = model_zoo.model_zoo.get_default_providers()
23 | self.session = model_zoo.model_zoo.PickableInferenceSession(
24 | model_file, providers=providers)
25 |
26 | self.input_mean = 127.5
27 | self.input_std = 127.5
28 | inputs = self.session.get_inputs()
29 | self.input_names = []
30 | for inp in inputs:
31 | self.input_names.append(inp.name)
32 | outputs = self.session.get_outputs()
33 | output_names = []
34 | for out in outputs:
35 | output_names.append(out.name)
36 | self.output_names = output_names
37 | assert len(
38 | self.output_names
39 | ) == 1, "The output number of PoseTransfer model should be 1, but got {}, please check your model.".format(
40 | len(self.output_names))
41 | output_shape = outputs[0].shape
42 | input_cfg = inputs[0]
43 | input_shape = input_cfg.shape
44 | self.input_shape = input_shape
45 | print('pose transfer shape:', self.input_shape)
46 |
47 | def forward(self, source_image, target_image, image_format='rgb'):
48 | h, w, c = source_image.shape
49 | if image_format == 'rgb':
50 | pass
51 | elif image_format == 'bgr':
52 | source_image = cv2.cvtColor(source_image, cv2.COLOR_BGR2RGB)
53 | target_image = cv2.cvtColor(target_image, cv2.COLOR_BGR2RGB)
54 | image_format = 'rgb'
55 | else:
56 | raise UserWarning(
57 | "PoseTransfer not support image format {}".format(image_format))
58 | imgA = self._resize_and_pad_image(source_image)
59 | kptA = self._estimate_keypoints(imgA, image_format=image_format)
60 | mapA = self._keypoints2heatmaps(kptA)
61 |
62 | imgB = self._resize_and_pad_image(target_image)
63 | kptB = self._estimate_keypoints(imgB)
64 | mapB = self._keypoints2heatmaps(kptB)
65 |
66 | imgA_t = (imgA.astype('float32') - self.input_mean) / self.input_std
67 | imgA_t = imgA_t.transpose([2, 0, 1])[None, ...]
68 | mapA_t = mapA.transpose([2, 0, 1])[None, ...]
69 | mapB_t = mapB.transpose([2, 0, 1])[None, ...]
70 | mapAB_t = np.concatenate((mapA_t, mapB_t), axis=1)
71 | pred = self.session.run(self.output_names, {
72 | self.input_names[0]: imgA_t,
73 | self.input_names[1]: mapAB_t
74 | })[0]
75 | target_image = pred.transpose((0, 2, 3, 1))[0]
76 | bgr_target_image = np.clip(
77 | self.input_std * target_image + self.input_mean, 0,
78 | 255).astype(np.uint8)[:, :, ::-1]
79 | crop_size = (256,
80 | min((256 * target_image.shape[1] // target_image.shape[0]),
81 | 176))
82 | bgr_image = center_crop(bgr_target_image, crop_size)
83 | bgr_image = cv2.resize(bgr_image, (w, h), interpolation=cv2.INTER_CUBIC)
84 | return bgr_image
85 |
86 | def get(self, source_image, target_image, image_format='rgb'):
87 | return self.forward(source_image, target_image, image_format)
88 |
89 | def _resize_and_pad_image(self, image: np.ndarray, size=256):
90 | w = size * image.shape[1] // image.shape[0]
91 | w_box = min(w, size * 11 // 16)
92 | image = cv2.resize(image, (w, size), interpolation=cv2.INTER_CUBIC)
93 | image = center_crop(image, (size, w_box))
94 | image = pad(image,
95 | size - w_box,
96 | size - w_box,
97 | size - w_box,
98 | size - w_box,
99 | fill=255)
100 | image = center_crop(image, (size, size))
101 | return image
102 |
103 | def _estimate_keypoints(self, image: np.ndarray, image_format='rgb'):
104 | keypoints = self.pose_estimator.get(image, image_format)
105 | keypoints = keypoints[0] if len(keypoints) > 0 else np.zeros(
106 | (18, 3), dtype=np.int32)
107 | keypoints[np.where(keypoints[:, 2] == 0), :2] = -1
108 | keypoints = keypoints[:, :2]
109 | return keypoints
110 |
111 | def _keypoints2heatmaps(self, keypoints, size=256):
112 | heatmaps = np.zeros((size, size, keypoints.shape[0]), dtype=np.float32)
113 | for k in range(keypoints.shape[0]):
114 | x, y = keypoints[k]
115 | if x == -1 or y == -1:
116 | continue
117 | heatmaps[y, x, k] = 1.0
118 | return heatmaps
119 |
--------------------------------------------------------------------------------
/dofaker/pose/pose_utils.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 |
3 |
4 | def _pad_image(image, stride=1, padvalue=0):
5 | assert len(image.shape) == 2 or len(image.shape) == 3
6 | h, w = image.shape[:2]
7 | pads = [None] * 4
8 | pads[0] = 0 # left
9 | pads[1] = 0 # top
10 | pads[2] = 0 if (w % stride == 0) else stride - (w % stride) # right
11 | pads[3] = 0 if (h % stride == 0) else stride - (h % stride) # bottom
12 | num_channels = 1 if len(image.shape) == 2 else image.shape[2]
13 | image_padded = np.ones(
14 | (h + pads[3], w + pads[2], num_channels), dtype=np.uint8) * padvalue
15 | image_padded = np.squeeze(image_padded)
16 | image_padded[:h, :w] = image
17 | return image_padded, pads
18 |
19 |
20 | def _get_keypoints(candidates, subsets):
21 | k = subsets.shape[0]
22 | keypoints = np.zeros((k, 18, 3), dtype=np.int32)
23 | for i in range(k):
24 | for j in range(18):
25 | index = np.int32(subsets[i][j])
26 | if index != -1:
27 | x, y = np.int32(candidates[index][:2])
28 | keypoints[i][j] = (x, y, 1)
29 | return keypoints
30 |
--------------------------------------------------------------------------------
/dofaker/pose_core.py:
--------------------------------------------------------------------------------
1 | import os
2 | import cv2
3 |
4 | import numpy as np
5 | from moviepy.editor import VideoFileClip
6 |
7 | from .pose import PoseEstimator, PoseTransfer
8 | from .face_enhance import GFPGAN
9 | from .super_resolution import BSRGAN
10 | from .face_det import FaceAnalysis
11 |
12 |
13 | class PoseSwapper:
14 |
15 | def __init__(self,
16 | pose_estimator_name='openpose_body',
17 | pose_estimator_model_dir='weights/models',
18 | pose_transfer_name='pose_transfer',
19 | pose_transfer_model_dir='weights/models',
20 | image_sr_model='bsrgan',
21 | image_sr_model_dir='weights/models',
22 | face_enhance_name='gfpgan',
23 | face_enhance_model_dir='weights/models/',
24 | face_det_model='buffalo_l',
25 | face_det_model_dir='weights/models',
26 | log_iters=10,
27 | use_enhancer=True,
28 | use_sr=True,
29 | scale=1):
30 | pose_estimator = PoseEstimator(name=pose_estimator_name,
31 | root=pose_estimator_model_dir)
32 | self.pose_transfer = PoseTransfer(name=pose_transfer_name,
33 | root=pose_transfer_model_dir,
34 | pose_estimator=pose_estimator)
35 |
36 | if use_enhancer:
37 | self.det_model = FaceAnalysis(name=face_det_model,
38 | root=face_det_model_dir)
39 | self.det_model.prepare(ctx_id=1, det_size=(640, 640))
40 | self.face_enhance = GFPGAN(name=face_enhance_name,
41 | root=face_enhance_model_dir)
42 | self.use_enhancer = use_enhancer
43 |
44 | if use_sr:
45 | self.sr_model = BSRGAN(name=image_sr_model,
46 | root=image_sr_model_dir,
47 | scale=scale)
48 | self.scale = scale
49 | else:
50 | self.scale = 1
51 | self.use_sr = use_sr
52 | self.log_iters = log_iters
53 |
54 | def run(self, input_path, target_path, output_dir='output'):
55 | assert os.path.exists(
56 | input_path), "The input path {} not exists.".format(input_path)
57 | assert os.path.exists(
58 | target_path), "The target path {} not exists.".format(target_path)
59 | os.makedirs(output_dir, exist_ok=True)
60 | assert input_path.lower().endswith(
61 | ('jpg', 'jpeg', 'webp', 'png', 'bmp')
62 | ), "pose swapper input must be image endswith ('jpg', 'jpeg', 'webp', 'png', 'bmp'), but got {}.".format(
63 | input_path)
64 | if target_path.lower().endswith(('jpg', 'jpeg', 'webp', 'png', 'bmp')):
65 | return self.transfer_image(input_path, target_path, output_dir)
66 | else:
67 | return self.transfer_video(input_path, target_path, output_dir)
68 |
69 | def transfer_image(self, input_path, target_path, output_dir):
70 | source = cv2.imread(input_path)
71 | target = cv2.imread(target_path)
72 | transferred_image = self.transfer_pose(source,
73 | target,
74 | image_format='bgr')
75 | base_name = os.path.basename(input_path)
76 | output_path = os.path.join(output_dir, base_name)
77 | cv2.imwrite(output_path, transferred_image)
78 | return output_path
79 |
80 | def transfer_video(self, input_path, target_path, output_dir):
81 | source = cv2.imread(input_path)
82 | video = cv2.VideoCapture(target_path)
83 | fps = video.get(cv2.CAP_PROP_FPS)
84 | total_frames = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
85 | height, width, _ = source.shape
86 | frame_size = (width, height)
87 | print('video fps: {}, total_frames: {}, width: {}, height: {}'.format(
88 | fps, total_frames, width, height))
89 |
90 | video_name = os.path.basename(input_path).split('.')[0]
91 | four_cc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
92 | temp_video_path = os.path.join(output_dir,
93 | 'temp_{}.mp4'.format(video_name))
94 | save_video_path = os.path.join(output_dir, '{}.mp4'.format(video_name))
95 | output_video = cv2.VideoWriter(
96 | temp_video_path, four_cc, fps,
97 | (int(frame_size[0] * self.scale), int(frame_size[1] * self.scale)))
98 | i = 0
99 | while video.isOpened():
100 | ret, frame = video.read()
101 | if ret:
102 | transferred_image = self.transfer_pose(source,
103 | frame,
104 | image_format='bgr')
105 | i += 1
106 | if i % self.log_iters == 0:
107 | print('processing {}/{}'.format(i, total_frames))
108 | output_video.write(transferred_image)
109 | else:
110 | break
111 |
112 | video.release()
113 | output_video.release()
114 | print(temp_video_path)
115 | self.add_audio_to_video(target_path, temp_video_path, save_video_path)
116 | os.remove(temp_video_path)
117 | return save_video_path
118 |
119 | def transfer_pose(self, source, target, image_format='bgr'):
120 | transferred_image = self.pose_transfer.get(source,
121 | target,
122 | image_format=image_format)
123 | if self.use_enhancer:
124 | faces = self.det_model.get(transferred_image, max_num=1)
125 | for face in faces:
126 | transferred_image = self.face_enhance.get(
127 | transferred_image,
128 | face,
129 | paste_back=True,
130 | image_format=image_format)
131 |
132 | if self.use_sr:
133 | transferred_image = self.sr_model.get(transferred_image,
134 | image_format=image_format)
135 | return transferred_image
136 |
137 | def add_audio_to_video(self, src_video_path, target_video_path,
138 | save_video_path):
139 | audio = VideoFileClip(src_video_path).audio
140 | target_video = VideoFileClip(target_video_path)
141 | target_video = target_video.set_audio(audio)
142 | target_video.write_videofile(save_video_path)
143 | return target_video_path
144 |
--------------------------------------------------------------------------------
/dofaker/super_resolution/__init__.py:
--------------------------------------------------------------------------------
1 | from .bsrgan import BSRGAN
2 |
--------------------------------------------------------------------------------
/dofaker/super_resolution/bsrgan.py:
--------------------------------------------------------------------------------
1 | import numpy as np
2 |
3 | import cv2
4 |
5 | from insightface import model_zoo
6 | from dofaker.utils import download_file, get_model_url
7 |
8 |
9 | class BSRGAN:
10 |
11 | def __init__(self, name='bsrgan', root='weights/models', scale=1) -> None:
12 | _, model_file = download_file(get_model_url(name),
13 | save_dir=root,
14 | overwrite=False)
15 | self.scale = scale
16 | providers = model_zoo.model_zoo.get_default_providers()
17 | self.session = model_zoo.model_zoo.PickableInferenceSession(
18 | model_file, providers=providers)
19 |
20 | self.input_mean = 0.0
21 | self.input_std = 255.0
22 | inputs = self.session.get_inputs()
23 | self.input_names = []
24 | for inp in inputs:
25 | self.input_names.append(inp.name)
26 | outputs = self.session.get_outputs()
27 | output_names = []
28 | for out in outputs:
29 | output_names.append(out.name)
30 | self.output_names = output_names
31 | assert len(
32 | self.output_names
33 | ) == 1, "The output number of BSRGAN model should be 1, but got {}, please check your model.".format(
34 | len(self.output_names))
35 | output_shape = outputs[0].shape
36 | input_cfg = inputs[0]
37 | input_shape = input_cfg.shape
38 | self.input_shape = input_shape
39 | print('image super resolution shape:', self.input_shape)
40 |
41 | def forward(self, image, image_format='bgr'):
42 | if isinstance(image, str):
43 | image = cv2.imread(image, 1)
44 | image_format = 'bgr'
45 | elif isinstance(image, np.ndarray):
46 | if image_format == 'bgr':
47 | image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
48 | elif image_format == 'rgb':
49 | pass
50 | else:
51 | raise UserWarning(
52 | "BSRGAN not support image format {}".format(image_format))
53 | else:
54 | raise UserWarning(
55 | "BSRGAN input must be str or np.ndarray, but got {}.".format(
56 | type(image)))
57 | img = (image - self.input_mean) / self.input_std
58 | pred = self.session.run(self.output_names,
59 | {self.input_names[0]: img})[0]
60 | return pred
61 |
62 | def get(self, img, image_format='bgr'):
63 | if image_format.lower() == 'bgr':
64 | img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
65 | elif image_format.lower() == 'rgb':
66 | pass
67 | else:
68 | raise UserWarning(
69 | "gfpgan not support image format {}".format(image_format))
70 | h, w, c = img.shape
71 | blob = cv2.dnn.blobFromImage(
72 | img,
73 | 1.0 / self.input_std, (w, h),
74 | (self.input_mean, self.input_mean, self.input_mean),
75 | swapRB=False)
76 | pred = self.session.run(self.output_names,
77 | {self.input_names[0]: blob})[0]
78 | image_aug = pred.transpose((0, 2, 3, 1))[0]
79 | rgb_aug = np.clip(self.input_std * image_aug + self.input_mean, 0,
80 | 255).astype(np.uint8)
81 | rgb_aug = cv2.resize(rgb_aug,
82 | (int(w * self.scale), int(h * self.scale)))
83 | bgr_aug = rgb_aug[:, :, ::-1]
84 | return bgr_aug
85 |
--------------------------------------------------------------------------------
/dofaker/transforms/__init__.py:
--------------------------------------------------------------------------------
1 | from .functional import center_crop, pad
2 |
--------------------------------------------------------------------------------
/dofaker/transforms/functional.py:
--------------------------------------------------------------------------------
1 | import numbers
2 | import numpy as np
3 |
4 | import cv2
5 |
6 |
7 | def center_crop(image: np.ndarray, output_size):
8 | if isinstance(output_size, numbers.Number):
9 | output_size = (int(output_size), int(output_size))
10 | elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
11 | output_size = (output_size[0], output_size[0])
12 |
13 | image_height, image_width, c = image.shape
14 | crop_height, crop_width = output_size
15 |
16 | if crop_width > image_width or crop_height > image_height:
17 | padding_ltrb = [
18 | (crop_width - image_width) // 2 if crop_width > image_width else 0,
19 | (crop_height - image_height) //
20 | 2 if crop_height > image_height else 0,
21 | (crop_width - image_width + 1) //
22 | 2 if crop_width > image_width else 0,
23 | (crop_height - image_height + 1) //
24 | 2 if crop_height > image_height else 0,
25 | ]
26 | image = cv2.copyMakeBorder(image,
27 | padding_ltrb[1],
28 | padding_ltrb[3],
29 | padding_ltrb[0],
30 | padding_ltrb[2],
31 | cv2.BORDER_CONSTANT,
32 | value=(0, 0, 0))
33 | image_height, image_width, c = image.shape
34 | if crop_width == image_width and crop_height == image_height:
35 | return image
36 |
37 | crop_top = int(round((image_height - crop_height) / 2.0))
38 | crop_left = int(round((image_width - crop_width) / 2.0))
39 | return image[crop_top:crop_top + crop_height,
40 | crop_left:crop_left + crop_width]
41 |
42 |
43 | def pad(image,
44 | left,
45 | top,
46 | right,
47 | bottom,
48 | fill: int = 0,
49 | padding_mode: str = "constant"):
50 | if padding_mode == 'constant':
51 | return cv2.copyMakeBorder(image,
52 | top,
53 | bottom,
54 | left,
55 | right,
56 | cv2.BORDER_CONSTANT,
57 | value=(fill, fill, fill))
58 | else:
59 | raise UserWarning('padding mode {} not supported.'.format(padding_mode))
60 |
--------------------------------------------------------------------------------
/dofaker/utils/__init__.py:
--------------------------------------------------------------------------------
1 | from .download import download_file
2 | from .weights_urls import get_model_url
3 |
--------------------------------------------------------------------------------
/dofaker/utils/download.py:
--------------------------------------------------------------------------------
1 | import os
2 | import requests
3 | import zipfile
4 |
5 | from tqdm import tqdm
6 |
7 |
8 | def download_file(url: str, save_dir='./', overwrite=False, unzip=True):
9 | os.makedirs(save_dir, exist_ok=True)
10 | file_name = url.split('/')[-1]
11 | file_path = os.path.join(save_dir, file_name)
12 |
13 | if os.path.exists(file_path) and not overwrite:
14 | pass
15 | else:
16 | print('Downloading file {} from {}...'.format(file_path, url))
17 |
18 | r = requests.get(url, stream=True)
19 | print(r.status_code)
20 | if r.status_code != 200:
21 | raise RuntimeError('Failed downloading url {}!'.format(url))
22 | total_length = r.headers.get('content-length')
23 | with open(file_path, 'wb') as f:
24 | if total_length is None: # no content length header
25 | for chunk in r.iter_content(chunk_size=1024):
26 | if chunk: # filter out keep-alive new chunks
27 | f.write(chunk)
28 | else:
29 | total_length = int(total_length)
30 | print('file length: ', int(total_length / 1024. + 0.5))
31 | for chunk in tqdm(r.iter_content(chunk_size=1024),
32 | total=int(total_length / 1024. + 0.5),
33 | unit='KB',
34 | unit_scale=False,
35 | dynamic_ncols=True):
36 | f.write(chunk)
37 | if unzip and file_path.endswith('.zip'):
38 | save_dir = file_path.split('.')[0]
39 | if os.path.isdir(save_dir) and os.path.exists(save_dir):
40 | pass
41 | else:
42 | with zipfile.ZipFile(file_path, 'r') as zip_ref:
43 | zip_ref.extractall(save_dir)
44 |
45 | return save_dir, file_path
46 |
--------------------------------------------------------------------------------
/dofaker/utils/utils.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/justld/dofaker/d81dc0c49f37f91430a34e20b5d3eab56ca58b9d/dofaker/utils/utils.py
--------------------------------------------------------------------------------
/dofaker/utils/weights_urls.py:
--------------------------------------------------------------------------------
1 | WEIGHT_URLS = {
2 | 'buffalo_l':
3 | 'https://github.com/justld/dofaker/releases/download/v0.1/buffalo_l.zip',
4 | 'buffalo_s':
5 | 'https://github.com/justld/dofaker/releases/download/v0.1/buffalo_s.zip',
6 | 'buffalo_sc':
7 | 'https://github.com/justld/dofaker/releases/download/v0.1/buffalo_sc.zip',
8 | 'inswapper':
9 | 'https://github.com/justld/dofaker/releases/download/v0.1/inswapper_128.onnx',
10 | 'gfpgan':
11 | 'https://github.com/justld/dofaker/releases/download/v0.1/GFPGANv1.3.onnx',
12 | 'bsrgan':
13 | 'https://github.com/justld/dofaker/releases/download/v0.1/bsrgan_4.onnx',
14 | 'openpose_body':
15 | 'https://github.com/justld/dofaker/releases/download/v0.1/openpose_body.onnx',
16 | 'pose_transfer':
17 | 'https://github.com/justld/dofaker/releases/download/v0.1/pose_transfer.onnx',
18 | }
19 |
20 |
21 | def get_model_url(model_name):
22 | return WEIGHT_URLS[model_name]
23 |
--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | moviepy
2 | gradio>=4.8
3 | insightface
4 | flake8
5 | yapf
6 |
--------------------------------------------------------------------------------
/run_faceswapper.py:
--------------------------------------------------------------------------------
1 | import argparse
2 | from dofaker import FaceSwapper
3 |
4 |
5 | def parse_args():
6 | parser = argparse.ArgumentParser(description='running face swap')
7 | parser.add_argument('--source',
8 | help='select an image or video to be swapped',
9 | dest='source',
10 | required=True)
11 | parser.add_argument('--dst_face_paths',
12 | help='select images in source to be swapped',
13 | dest='dst_face_paths',
14 | nargs='+',
15 | default=None)
16 | parser.add_argument(
17 | '--src_face_paths',
18 | help='select images to replace dst_faces in source image or video.',
19 | dest='src_face_paths',
20 | nargs='+',
21 | required=True)
22 | parser.add_argument('--output_dir',
23 | help='output directory',
24 | dest='output_dir',
25 | default='output')
26 | parser.add_argument('--det_model_name',
27 | help='detection model name for insightface',
28 | dest='det_model_name',
29 | default='buffalo_l')
30 | parser.add_argument('--det_model_dir',
31 | help='detection model dir for insightface',
32 | dest='det_model_dir',
33 | default='weights/models')
34 | parser.add_argument('--swap_model_name',
35 | help='swap model name',
36 | dest='swap_model_name',
37 | default='inswapper')
38 | parser.add_argument('--image_sr_model',
39 | help='image super resolution model',
40 | dest='image_sr_model',
41 | default='bsrgan')
42 | parser.add_argument('--face_swap_model_dir',
43 | help='swap model path',
44 | dest='face_swap_model_dir',
45 | default='weights/models')
46 | parser.add_argument('--image_sr_model_dir',
47 | help='image super resolution model dir',
48 | dest='image_sr_model_dir',
49 | default='weights/models')
50 | parser.add_argument('--face_enhance_name',
51 | help='face enhance model',
52 | dest='face_enhance_name',
53 | default='gfpgan')
54 | parser.add_argument('--face_enhance_model_dir',
55 | help='face enhance model dir',
56 | dest='face_enhance_model_dir',
57 | default='weights/models')
58 | parser.add_argument('--face_sim_thre',
59 | help='similarity of face embedding threshold',
60 | dest='face_sim_thre',
61 | default=0.5)
62 | parser.add_argument('--log_iters',
63 | help='print log intervals',
64 | dest='log_iters',
65 | default=10,
66 | type=int)
67 | parser.add_argument('--use_enhancer',
68 | help='whether use face enhance model',
69 | dest='use_enhancer',
70 | action='store_true')
71 | parser.add_argument('--use_sr',
72 | help='whether use image super resolution model',
73 | dest='use_sr',
74 | action='store_true')
75 | parser.add_argument('--sr_scale',
76 | help='image super resolution scale',
77 | dest='sr_scale',
78 | default=1,
79 | type=float)
80 | return parser.parse_args()
81 |
82 |
83 | if __name__ == '__main__':
84 | args = parse_args()
85 | faker = FaceSwapper(
86 | face_det_model=args.det_model_name,
87 | face_det_model_dir=args.det_model_dir,
88 | face_swap_model=args.swap_model_name,
89 | face_swap_model_dir=args.face_swap_model_dir,
90 | image_sr_model=args.image_sr_model,
91 | image_sr_model_dir=args.image_sr_model_dir,
92 | face_enhance_model=args.face_enhance_name,
93 | face_enhance_model_dir=args.face_enhance_model_dir,
94 | face_sim_thre=args.face_sim_thre,
95 | log_iters=args.log_iters,
96 | use_enhancer=args.use_enhancer,
97 | use_sr=args.use_sr,
98 | scale=args.sr_scale,
99 | )
100 |
101 | faker.run(
102 | input_path=args.source,
103 | dst_face_paths=args.dst_face_paths,
104 | src_face_paths=args.src_face_paths,
105 | output_dir=args.output_dir,
106 | )
107 |
--------------------------------------------------------------------------------
/run_posetransfer.py:
--------------------------------------------------------------------------------
1 | import argparse
2 | from dofaker import PoseSwapper
3 |
4 |
5 | def parse_args():
6 | parser = argparse.ArgumentParser(description='running face swap')
7 | parser.add_argument('--source',
8 | help='select an image or video to be swapped',
9 | dest='source',
10 | required=True)
11 | parser.add_argument('--target',
12 | help='the target pose image',
13 | dest='target',
14 | required=True)
15 | parser.add_argument('--output_dir',
16 | help='output directory',
17 | dest='output_dir',
18 | default='output')
19 | parser.add_argument('--pose_estimator_name',
20 | help='pose estimator name',
21 | dest='pose_estimator_name',
22 | default='openpose_body')
23 | parser.add_argument('--pose_estimator_model_dir',
24 | help='pose estimator model dir',
25 | dest='pose_estimator_model_dir',
26 | default='weights/models')
27 | parser.add_argument('--pose_transfer_name',
28 | help='pose transfer name',
29 | dest='pose_transfer_name',
30 | default='pose_transfer')
31 | parser.add_argument('--pose_transfer_model_dir',
32 | help='pose transfer model dir',
33 | dest='pose_transfer_model_dir',
34 | default='weights/models')
35 | parser.add_argument('--det_model_name',
36 | help='detection model name for insightface',
37 | dest='det_model_name',
38 | default='buffalo_l')
39 | parser.add_argument('--det_model_dir',
40 | help='detection model dir for insightface',
41 | dest='det_model_dir',
42 | default='weights/models')
43 | parser.add_argument('--image_sr_model',
44 | help='image super resolution model',
45 | dest='image_sr_model',
46 | default='bsrgan')
47 | parser.add_argument('--image_sr_model_dir',
48 | help='image super resolution model dir',
49 | dest='image_sr_model_dir',
50 | default='weights/models')
51 | parser.add_argument('--face_enhance_name',
52 | help='face enhance model',
53 | dest='face_enhance_name',
54 | default='gfpgan')
55 | parser.add_argument('--face_enhance_model_dir',
56 | help='face enhance model dir',
57 | dest='face_enhance_model_dir',
58 | default='weights/models')
59 | parser.add_argument('--log_iters',
60 | help='print log intervals',
61 | dest='log_iters',
62 | default=10,
63 | type=int)
64 | parser.add_argument('--use_enhancer',
65 | help='whether use face enhance model',
66 | dest='use_enhancer',
67 | action='store_true')
68 | parser.add_argument('--use_sr',
69 | help='whether use image super resolution model',
70 | dest='use_sr',
71 | action='store_true')
72 | parser.add_argument('--sr_scale',
73 | help='image super resolution scale',
74 | dest='sr_scale',
75 | default=1,
76 | type=float)
77 | return parser.parse_args()
78 |
79 |
80 | if __name__ == '__main__':
81 | args = parse_args()
82 | faker = PoseSwapper(
83 | pose_estimator_name=args.pose_estimator_name,
84 | pose_estimator_model_dir=args.pose_estimator_model_dir,
85 | pose_transfer_name=args.pose_transfer_name,
86 | pose_transfer_model_dir=args.pose_transfer_model_dir,
87 | face_det_model=args.det_model_name,
88 | face_det_model_dir=args.det_model_dir,
89 | image_sr_model=args.image_sr_model,
90 | image_sr_model_dir=args.image_sr_model_dir,
91 | face_enhance_name=args.face_enhance_name,
92 | face_enhance_model_dir=args.face_enhance_model_dir,
93 | log_iters=args.log_iters,
94 | use_enhancer=args.use_enhancer,
95 | use_sr=args.use_sr,
96 | scale=args.sr_scale,
97 | )
98 |
99 | faker.run(
100 | input_path=args.source,
101 | target_path=args.target,
102 | output_dir=args.output_dir,
103 | )
104 |
--------------------------------------------------------------------------------
/setup.py:
--------------------------------------------------------------------------------
1 | from setuptools import setup, find_packages
2 |
3 | with open('requirements.txt') as file:
4 | REQUIRED_PACKAGES = file.read()
5 |
6 | setup(name='dofaker',
7 | version='0.1',
8 | keywords=('face swap'),
9 | description='A simple face swap tool',
10 | url='https://github.com/justld/dofaker',
11 | author='justld',
12 | author_email='1207540056@qq.com',
13 | packages=find_packages(),
14 | include_package_data=True,
15 | platforms='any',
16 | install_requires=REQUIRED_PACKAGES,
17 | scripts=[],
18 | license='GPL 3.0',
19 | entry_points={'console_scripts': [
20 | 'dofaker = web_ui:main',
21 | ]})
22 |
--------------------------------------------------------------------------------
/test.sh:
--------------------------------------------------------------------------------
1 | # python run_faceswapper.py --source docs/test/multi.png --dst_face_paths docs/test/dst1.png --src_face_paths docs/test/trump.jpg
2 |
3 | # python run_faceswapper.py --source docs/test/multi.png --dst_face_paths docs/test/dst1.png docs/test/dst2.png --src_face_paths docs/test/trump.jpg docs/test/taitan.jpeg
4 |
5 | # python run_faceswapper.py --source docs/test/multi.png --dst_face_paths docs/test/dst1.png docs/test/dst2.png --src_face_paths docs/test/trump.jpg docs/test/taitan.jpeg --use_enhancer
6 |
7 | python run_faceswapper.py --source docs/test/multi.png --dst_face_paths docs/test/dst1.png docs/test/dst2.png --src_face_paths docs/test/trump.jpg docs/test/taitan.jpeg --use_enhancer --use_sr --sr_scale 2.0
8 |
9 | # python run_posetransfer.py --source docs/test/condition.jpg --target docs/test/target_pose_reference.jpg
10 |
11 | # python run_posetransfer.py --source docs/test/condition.jpg --target docs/test/target_pose_reference.jpg --use_enhancer
12 |
13 | python run_posetransfer.py --source docs/test/condition.jpg --target docs/test/target_pose_reference.jpg --use_enhancer --use_sr --sr_scale 2.0
14 |
--------------------------------------------------------------------------------
/web_ui.py:
--------------------------------------------------------------------------------
1 | import argparse
2 |
3 | import gradio as gr
4 | from dofaker import FaceSwapper, PoseSwapper
5 |
6 |
7 | def parse_args():
8 | parser = argparse.ArgumentParser(description='running face swap')
9 | parser.add_argument(
10 | '--inbrowser',
11 | help=
12 | 'whether to automatically launch the interface in a new tab on the default browser.',
13 | dest='inbrowser',
14 | default=True)
15 | parser.add_argument(
16 | '--server_port',
17 | help=
18 | 'will start gradio app on this port (if available). Can be set by environment variable GRADIO_SERVER_PORT. If None, will search for an available port starting at 7860.',
19 | dest='server_port',
20 | type=int,
21 | default=None)
22 | return parser.parse_args()
23 |
24 |
25 | def swap_face(input_path, dst_path, src_path, use_enhancer, use_sr, scale,
26 | face_sim_thre):
27 | faker = FaceSwapper(use_enhancer=use_enhancer,
28 | use_sr=use_sr,
29 | scale=scale,
30 | face_sim_thre=face_sim_thre)
31 | output_path = faker.run(input_path, dst_path, src_path)
32 | return output_path
33 |
34 |
35 | def swap_pose(input_path, target_path, use_enhancer, use_sr, scale):
36 | faker = PoseSwapper(use_enhancer=use_enhancer, use_sr=use_sr, scale=scale)
37 | output_path = faker.run(input_path, target_path)
38 | return output_path
39 |
40 |
41 | def main():
42 | args = parse_args()
43 |
44 | with gr.Blocks(title='DoFaker') as web_ui:
45 | gr.Markdown('DoFaker: Face Swap and pose swap web ui')
46 | with gr.Tab('FaceSwapper'):
47 | gr.Markdown('DoFaker: Face Swap Web UI')
48 | with gr.Tab('Image'):
49 | with gr.Row():
50 | with gr.Column():
51 | gr.Markdown('The source image to be swapped')
52 | image_input = gr.Image(type='filepath')
53 | with gr.Row():
54 | with gr.Column():
55 | gr.Markdown(
56 | 'target face included in source image')
57 | dst_face_image = gr.Image(type='filepath')
58 | with gr.Column():
59 | gr.Markdown(
60 | 'source face to replace target face')
61 | src_face_image = gr.Image(type='filepath')
62 |
63 | with gr.Column():
64 | output_image = gr.Image(type='filepath')
65 | use_enhancer = gr.Checkbox(
66 | label="face enhance",
67 | info="Whether use face enhance model.")
68 | with gr.Row():
69 | use_sr = gr.Checkbox(
70 | label="super resolution",
71 | info="Whether use image resolution model.")
72 | scale = gr.Number(
73 | value=1, label='image super resolution scale')
74 | with gr.Row():
75 | face_sim_thre = gr.Number(
76 | value=0.6,
77 | label='face similarity threshold',
78 | minimum=0.0,
79 | maximum=1.0)
80 | convert_button = gr.Button('Swap')
81 | convert_button.click(fn=swap_face,
82 | inputs=[
83 | image_input, dst_face_image,
84 | src_face_image, use_enhancer,
85 | use_sr, scale, face_sim_thre
86 | ],
87 | outputs=[output_image],
88 | api_name='image swap')
89 |
90 | with gr.Tab('Video'):
91 | with gr.Row():
92 | with gr.Column():
93 | gr.Markdown('The source video to be swapped')
94 | video_input = gr.Video()
95 | with gr.Row():
96 | with gr.Column():
97 | gr.Markdown(
98 | 'target face included in source image')
99 | dst_face_image = gr.Image(type='filepath')
100 | with gr.Column():
101 | gr.Markdown(
102 | 'source face to replace target face')
103 | src_face_image = gr.Image(type='filepath')
104 |
105 | with gr.Column():
106 | output_video = gr.Video()
107 | use_enhancer = gr.Checkbox(
108 | label="face enhance",
109 | info="Whether use face enhance model.")
110 | with gr.Row():
111 | use_sr = gr.Checkbox(
112 | label="super resolution",
113 | info="Whether use image resolution model.")
114 | scale = gr.Number(
115 | value=1, label='image super resolution scale')
116 | with gr.Row():
117 | face_sim_thre = gr.Number(
118 | value=0.6,
119 | label='face similarity threshold',
120 | minimum=0.0,
121 | maximum=1.0)
122 | convert_button = gr.Button('Swap')
123 | convert_button.click(fn=swap_face,
124 | inputs=[
125 | video_input, dst_face_image,
126 | src_face_image, use_enhancer,
127 | use_sr, scale, face_sim_thre
128 | ],
129 | outputs=[output_video],
130 | api_name='video swap')
131 |
132 | with gr.Tab('PoseSwapper'):
133 | gr.Markdown('DoFaker: Pose Swap Web UI')
134 | with gr.Tab('Image'):
135 | with gr.Row():
136 | with gr.Column():
137 | gr.Markdown('The source image to be swapped')
138 | image_input = gr.Image(type='filepath')
139 | gr.Markdown('The target image with pose')
140 | target = gr.Image(type='filepath')
141 |
142 | with gr.Column():
143 | output_image = gr.Image(type='filepath')
144 | use_enhancer = gr.Checkbox(
145 | label="face enhance",
146 | info="Whether use face enhance model.")
147 | with gr.Row():
148 | use_sr = gr.Checkbox(
149 | label="super resolution",
150 | info="Whether use image resolution model.")
151 | scale = gr.Number(
152 | value=1, label='image super resolution scale')
153 | convert_button = gr.Button('Swap')
154 | convert_button.click(fn=swap_pose,
155 | inputs=[
156 | image_input, target,
157 | use_enhancer, use_sr, scale
158 | ],
159 | outputs=[output_image],
160 | api_name='image swap')
161 |
162 | # with gr.Tab('Video'):
163 | # with gr.Row():
164 | # with gr.Column():
165 | # gr.Markdown('The source video to be swapped')
166 | # video_input = gr.Image(type='filepath')
167 | # gr.Markdown('The target image with pose')
168 | # target = gr.Video()
169 |
170 | # with gr.Column():
171 | # output_video = gr.Video()
172 | # use_enhancer = gr.Checkbox(
173 | # label="face enhance",
174 | # info="Whether use face enhance model.")
175 | # with gr.Row():
176 | # use_sr = gr.Checkbox(
177 | # label="super resolution",
178 | # info="Whether use image resolution model.")
179 | # scale = gr.Number(value=1,
180 | # label='image super resolution scale')
181 | # convert_button = gr.Button('Swap')
182 | # convert_button.click(fn=swap_pose,
183 | # inputs=[
184 | # video_input, target, use_enhancer,
185 | # use_sr, scale
186 | # ],
187 | # outputs=[output_video],
188 | # api_name='video swap')
189 |
190 | web_ui.launch(inbrowser=args.inbrowser, server_port=args.server_port)
191 |
192 |
193 | if __name__ == '__main__':
194 | main()
195 |
--------------------------------------------------------------------------------