puirjvfyc
/
style-transfer-app
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

first commit

Browse Source
master
Yudong An 2 years ago
parent e501108fa0
commit 96cd7bc678
24 changed files with 12324 additions and 1 deletions
Show Stats Download Patch File Download Diff File

674
LICENSE
Unescape View File

@ -0,0 +1,674 @@
GNU GENERAL PUBLIC LICENSE
Version 3, 29 June 2007
Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
Everyone is permitted to copy and distribute verbatim copies
of this license document, but changing it is not allowed.
Preamble
The GNU General Public License is a free, copyleft license for
software and other kinds of works.
The licenses for most software and other practical works are designed
to take away your freedom to share and change the works. By contrast,
the GNU General Public License is intended to guarantee your freedom to
share and change all versions of a program--to make sure it remains free
software for all its users. We, the Free Software Foundation, use the
GNU General Public License for most of our software; it applies also to
any other work released this way by its authors. You can apply it to
your programs, too.
When we speak of free software, we are referring to freedom, not
price. Our General Public Licenses are designed to make sure that you
have the freedom to distribute copies of free software (and charge for
them if you wish), that you receive source code or can get it if you
want it, that you can change the software or use pieces of it in new
free programs, and that you know you can do these things.
To protect your rights, we need to prevent others from denying you
these rights or asking you to surrender the rights. Therefore, you have
certain responsibilities if you distribute copies of the software, or if
you modify it: responsibilities to respect the freedom of others.
For example, if you distribute copies of such a program, whether
gratis or for a fee, you must pass on to the recipients the same
freedoms that you received. You must make sure that they, too, receive
or can get the source code. And you must show them these terms so they
know their rights.
Developers that use the GNU GPL protect your rights with two steps:
(1) assert copyright on the software, and (2) offer you this License
giving you legal permission to copy, distribute and/or modify it.
For the developers' and authors' protection, the GPL clearly explains
that there is no warranty for this free software. For both users' and
authors' sake, the GPL requires that modified versions be marked as
changed, so that their problems will not be attributed erroneously to
authors of previous versions.
Some devices are designed to deny users access to install or run
modified versions of the software inside them, although the manufacturer
can do so. This is fundamentally incompatible with the aim of
protecting users' freedom to change the software. The systematic
pattern of such abuse occurs in the area of products for individuals to
use, which is precisely where it is most unacceptable. Therefore, we
have designed this version of the GPL to prohibit the practice for those
products. If such problems arise substantially in other domains, we
stand ready to extend this provision to those domains in future versions
of the GPL, as needed to protect the freedom of users.
Finally, every program is threatened constantly by software patents.
States should not allow patents to restrict development and use of
software on general-purpose computers, but in those that do, we wish to
avoid the special danger that patents applied to a free program could
make it effectively proprietary. To prevent this, the GPL assures that
patents cannot be used to render the program non-free.
The precise terms and conditions for copying, distribution and
modification follow.
TERMS AND CONDITIONS
0. Definitions.
"This License" refers to version 3 of the GNU General Public License.
"Copyright" also means copyright-like laws that apply to other kinds of
works, such as semiconductor masks.
"The Program" refers to any copyrightable work licensed under this
License. Each licensee is addressed as "you". "Licensees" and
"recipients" may be individuals or organizations.
To "modify" a work means to copy from or adapt all or part of the work
in a fashion requiring copyright permission, other than the making of an
exact copy. The resulting work is called a "modified version" of the
earlier work or a work "based on" the earlier work.
A "covered work" means either the unmodified Program or a work based
on the Program.
To "propagate" a work means to do anything with it that, without
permission, would make you directly or secondarily liable for
infringement under applicable copyright law, except executing it on a
computer or modifying a private copy. Propagation includes copying,
distribution (with or without modification), making available to the
public, and in some countries other activities as well.
To "convey" a work means any kind of propagation that enables other
parties to make or receive copies. Mere interaction with a user through
a computer network, with no transfer of a copy, is not conveying.
An interactive user interface displays "Appropriate Legal Notices"
to the extent that it includes a convenient and prominently visible
feature that (1) displays an appropriate copyright notice, and (2)
tells the user that there is no warranty for the work (except to the
extent that warranties are provided), that licensees may convey the
work under this License, and how to view a copy of this License. If
the interface presents a list of user commands or options, such as a
menu, a prominent item in the list meets this criterion.
1. Source Code.
The "source code" for a work means the preferred form of the work
for making modifications to it. "Object code" means any non-source
form of a work.
A "Standard Interface" means an interface that either is an official
standard defined by a recognized standards body, or, in the case of
interfaces specified for a particular programming language, one that
is widely used among developers working in that language.
The "System Libraries" of an executable work include anything, other
than the work as a whole, that (a) is included in the normal form of
packaging a Major Component, but which is not part of that Major
Component, and (b) serves only to enable use of the work with that
Major Component, or to implement a Standard Interface for which an
implementation is available to the public in source code form. A
"Major Component", in this context, means a major essential component
(kernel, window system, and so on) of the specific operating system
(if any) on which the executable work runs, or a compiler used to
produce the work, or an object code interpreter used to run it.
The "Corresponding Source" for a work in object code form means all
the source code needed to generate, install, and (for an executable
work) run the object code and to modify the work, including scripts to
control those activities. However, it does not include the work's
System Libraries, or general-purpose tools or generally available free
programs which are used unmodified in performing those activities but
which are not part of the work. For example, Corresponding Source
includes interface definition files associated with source files for
the work, and the source code for shared libraries and dynamically
linked subprograms that the work is specifically designed to require,
such as by intimate data communication or control flow between those
subprograms and other parts of the work.
The Corresponding Source need not include anything that users
can regenerate automatically from other parts of the Corresponding
Source.
The Corresponding Source for a work in source code form is that
same work.
2. Basic Permissions.
All rights granted under this License are granted for the term of
copyright on the Program, and are irrevocable provided the stated
conditions are met. This License explicitly affirms your unlimited
permission to run the unmodified Program. The output from running a
covered work is covered by this License only if the output, given its
content, constitutes a covered work. This License acknowledges your
rights of fair use or other equivalent, as provided by copyright law.
You may make, run and propagate covered works that you do not
convey, without conditions so long as your license otherwise remains
in force. You may convey covered works to others for the sole purpose
of having them make modifications exclusively for you, or provide you
with facilities for running those works, provided that you comply with
the terms of this License in conveying all material for which you do
not control copyright. Those thus making or running the covered works
for you must do so exclusively on your behalf, under your direction
and control, on terms that prohibit them from making any copies of
your copyrighted material outside their relationship with you.
Conveying under any other circumstances is permitted solely under
the conditions stated below. Sublicensing is not allowed; section 10
makes it unnecessary.
3. Protecting Users' Legal Rights From Anti-Circumvention Law.
No covered work shall be deemed part of an effective technological
measure under any applicable law fulfilling obligations under article
11 of the WIPO copyright treaty adopted on 20 December 1996, or
similar laws prohibiting or restricting circumvention of such
measures.
When you convey a covered work, you waive any legal power to forbid
circumvention of technological measures to the extent such circumvention
is effected by exercising rights under this License with respect to
the covered work, and you disclaim any intention to limit operation or
modification of the work as a means of enforcing, against the work's
users, your or third parties' legal rights to forbid circumvention of
technological measures.
4. Conveying Verbatim Copies.
You may convey verbatim copies of the Program's source code as you
receive it, in any medium, provided that you conspicuously and
appropriately publish on each copy an appropriate copyright notice;
keep intact all notices stating that this License and any
non-permissive terms added in accord with section 7 apply to the code;
keep intact all notices of the absence of any warranty; and give all
recipients a copy of this License along with the Program.
You may charge any price or no price for each copy that you convey,
and you may offer support or warranty protection for a fee.
5. Conveying Modified Source Versions.
You may convey a work based on the Program, or the modifications to
produce it from the Program, in the form of source code under the
terms of section 4, provided that you also meet all of these conditions:
a) The work must carry prominent notices stating that you modified
it, and giving a relevant date.
b) The work must carry prominent notices stating that it is
released under this License and any conditions added under section
7. This requirement modifies the requirement in section 4 to
"keep intact all notices".
c) You must license the entire work, as a whole, under this
License to anyone who comes into possession of a copy. This
License will therefore apply, along with any applicable section 7
additional terms, to the whole of the work, and all its parts,
regardless of how they are packaged. This License gives no
permission to license the work in any other way, but it does not
invalidate such permission if you have separately received it.
d) If the work has interactive user interfaces, each must display
Appropriate Legal Notices; however, if the Program has interactive
interfaces that do not display Appropriate Legal Notices, your
work need not make them do so.
A compilation of a covered work with other separate and independent
works, which are not by their nature extensions of the covered work,
and which are not combined with it such as to form a larger program,
in or on a volume of a storage or distribution medium, is called an
"aggregate" if the compilation and its resulting copyright are not
used to limit the access or legal rights of the compilation's users
beyond what the individual works permit. Inclusion of a covered work
in an aggregate does not cause this License to apply to the other
parts of the aggregate.
6. Conveying Non-Source Forms.
You may convey a covered work in object code form under the terms
of sections 4 and 5, provided that you also convey the
machine-readable Corresponding Source under the terms of this License,
in one of these ways:
a) Convey the object code in, or embodied in, a physical product
(including a physical distribution medium), accompanied by the
Corresponding Source fixed on a durable physical medium
customarily used for software interchange.
b) Convey the object code in, or embodied in, a physical product
(including a physical distribution medium), accompanied by a
written offer, valid for at least three years and valid for as
long as you offer spare parts or customer support for that product
model, to give anyone who possesses the object code either (1) a
copy of the Corresponding Source for all the software in the
product that is covered by this License, on a durable physical
medium customarily used for software interchange, for a price no
more than your reasonable cost of physically performing this
conveying of source, or (2) access to copy the
Corresponding Source from a network server at no charge.
c) Convey individual copies of the object code with a copy of the
written offer to provide the Corresponding Source. This
alternative is allowed only occasionally and noncommercially, and
only if you received the object code with such an offer, in accord
with subsection 6b.
d) Convey the object code by offering access from a designated
place (gratis or for a charge), and offer equivalent access to the
Corresponding Source in the same way through the same place at no
further charge. You need not require recipients to copy the
Corresponding Source along with the object code. If the place to
copy the object code is a network server, the Corresponding Source
may be on a different server (operated by you or a third party)
that supports equivalent copying facilities, provided you maintain
clear directions next to the object code saying where to find the
Corresponding Source. Regardless of what server hosts the
Corresponding Source, you remain obligated to ensure that it is
available for as long as needed to satisfy these requirements.
e) Convey the object code using peer-to-peer transmission, provided
you inform other peers where the object code and Corresponding
Source of the work are being offered to the general public at no
charge under subsection 6d.
A separable portion of the object code, whose source code is excluded
from the Corresponding Source as a System Library, need not be
included in conveying the object code work.
A "User Product" is either (1) a "consumer product", which means any
tangible personal property which is normally used for personal, family,
or household purposes, or (2) anything designed or sold for incorporation
into a dwelling. In determining whether a product is a consumer product,
doubtful cases shall be resolved in favor of coverage. For a particular
product received by a particular user, "normally used" refers to a
typical or common use of that class of product, regardless of the status
of the particular user or of the way in which the particular user
actually uses, or expects or is expected to use, the product. A product
is a consumer product regardless of whether the product has substantial
commercial, industrial or non-consumer uses, unless such uses represent
the only significant mode of use of the product.
"Installation Information" for a User Product means any methods,
procedures, authorization keys, or other information required to install
and execute modified versions of a covered work in that User Product from
a modified version of its Corresponding Source. The information must
suffice to ensure that the continued functioning of the modified object
code is in no case prevented or interfered with solely because
modification has been made.
If you convey an object code work under this section in, or with, or
specifically for use in, a User Product, and the conveying occurs as
part of a transaction in which the right of possession and use of the
User Product is transferred to the recipient in perpetuity or for a
fixed term (regardless of how the transaction is characterized), the
Corresponding Source conveyed under this section must be accompanied
by the Installation Information. But this requirement does not apply
if neither you nor any third party retains the ability to install
modified object code on the User Product (for example, the work has
been installed in ROM).
The requirement to provide Installation Information does not include a
requirement to continue to provide support service, warranty, or updates
for a work that has been modified or installed by the recipient, or for
the User Product in which it has been modified or installed. Access to a
network may be denied when the modification itself materially and
adversely affects the operation of the network or violates the rules and
protocols for communication across the network.
Corresponding Source conveyed, and Installation Information provided,
in accord with this section must be in a format that is publicly
documented (and with an implementation available to the public in
source code form), and must require no special password or key for
unpacking, reading or copying.
7. Additional Terms.
"Additional permissions" are terms that supplement the terms of this
License by making exceptions from one or more of its conditions.
Additional permissions that are applicable to the entire Program shall
be treated as though they were included in this License, to the extent
that they are valid under applicable law. If additional permissions
apply only to part of the Program, that part may be used separately
under those permissions, but the entire Program remains governed by
this License without regard to the additional permissions.
When you convey a copy of a covered work, you may at your option
remove any additional permissions from that copy, or from any part of
it. (Additional permissions may be written to require their own
removal in certain cases when you modify the work.) You may place
additional permissions on material, added by you to a covered work,
for which you have or can give appropriate copyright permission.
Notwithstanding any other provision of this License, for material you
add to a covered work, you may (if authorized by the copyright holders of
that material) supplement the terms of this License with terms:
a) Disclaiming warranty or limiting liability differently from the
terms of sections 15 and 16 of this License; or
b) Requiring preservation of specified reasonable legal notices or
author attributions in that material or in the Appropriate Legal
Notices displayed by works containing it; or
c) Prohibiting misrepresentation of the origin of that material, or
requiring that modified versions of such material be marked in
reasonable ways as different from the original version; or
d) Limiting the use for publicity purposes of names of licensors or
authors of the material; or
e) Declining to grant rights under trademark law for use of some
trade names, trademarks, or service marks; or
f) Requiring indemnification of licensors and authors of that
material by anyone who conveys the material (or modified versions of
it) with contractual assumptions of liability to the recipient, for
any liability that these contractual assumptions directly impose on
those licensors and authors.
All other non-permissive additional terms are considered "further
restrictions" within the meaning of section 10. If the Program as you
received it, or any part of it, contains a notice stating that it is
governed by this License along with a term that is a further
restriction, you may remove that term. If a license document contains
a further restriction but permits relicensing or conveying under this
License, you may add to a covered work material governed by the terms
of that license document, provided that the further restriction does
not survive such relicensing or conveying.
If you add terms to a covered work in accord with this section, you
must place, in the relevant source files, a statement of the
additional terms that apply to those files, or a notice indicating
where to find the applicable terms.
Additional terms, permissive or non-permissive, may be stated in the
form of a separately written license, or stated as exceptions;
the above requirements apply either way.
8. Termination.
You may not propagate or modify a covered work except as expressly
provided under this License. Any attempt otherwise to propagate or
modify it is void, and will automatically terminate your rights under
this License (including any patent licenses granted under the third
paragraph of section 11).
However, if you cease all violation of this License, then your
license from a particular copyright holder is reinstated (a)
provisionally, unless and until the copyright holder explicitly and
finally terminates your license, and (b) permanently, if the copyright
holder fails to notify you of the violation by some reasonable means
prior to 60 days after the cessation.
Moreover, your license from a particular copyright holder is
reinstated permanently if the copyright holder notifies you of the
violation by some reasonable means, this is the first time you have
received notice of violation of this License (for any work) from that
copyright holder, and you cure the violation prior to 30 days after
your receipt of the notice.
Termination of your rights under this section does not terminate the
licenses of parties who have received copies or rights from you under
this License. If your rights have been terminated and not permanently
reinstated, you do not qualify to receive new licenses for the same
material under section 10.
9. Acceptance Not Required for Having Copies.
You are not required to accept this License in order to receive or
run a copy of the Program. Ancillary propagation of a covered work
occurring solely as a consequence of using peer-to-peer transmission
to receive a copy likewise does not require acceptance. However,
nothing other than this License grants you permission to propagate or
modify any covered work. These actions infringe copyright if you do
not accept this License. Therefore, by modifying or propagating a
covered work, you indicate your acceptance of this License to do so.
10. Automatic Licensing of Downstream Recipients.
Each time you convey a covered work, the recipient automatically
receives a license from the original licensors, to run, modify and
propagate that work, subject to this License. You are not responsible
for enforcing compliance by third parties with this License.
An "entity transaction" is a transaction transferring control of an
organization, or substantially all assets of one, or subdividing an
organization, or merging organizations. If propagation of a covered
work results from an entity transaction, each party to that
transaction who receives a copy of the work also receives whatever
licenses to the work the party's predecessor in interest had or could
give under the previous paragraph, plus a right to possession of the
Corresponding Source of the work from the predecessor in interest, if
the predecessor has it or can get it with reasonable efforts.
You may not impose any further restrictions on the exercise of the
rights granted or affirmed under this License. For example, you may
not impose a license fee, royalty, or other charge for exercise of
rights granted under this License, and you may not initiate litigation
(including a cross-claim or counterclaim in a lawsuit) alleging that
any patent claim is infringed by making, using, selling, offering for
sale, or importing the Program or any portion of it.
11. Patents.
A "contributor" is a copyright holder who authorizes use under this
License of the Program or a work on which the Program is based. The
work thus licensed is called the contributor's "contributor version".
A contributor's "essential patent claims" are all patent claims
owned or controlled by the contributor, whether already acquired or
hereafter acquired, that would be infringed by some manner, permitted
by this License, of making, using, or selling its contributor version,
but do not include claims that would be infringed only as a
consequence of further modification of the contributor version. For
purposes of this definition, "control" includes the right to grant
patent sublicenses in a manner consistent with the requirements of
this License.
Each contributor grants you a non-exclusive, worldwide, royalty-free
patent license under the contributor's essential patent claims, to
make, use, sell, offer for sale, import and otherwise run, modify and
propagate the contents of its contributor version.
In the following three paragraphs, a "patent license" is any express
agreement or commitment, however denominated, not to enforce a patent
(such as an express permission to practice a patent or covenant not to
sue for patent infringement). To "grant" such a patent license to a
party means to make such an agreement or commitment not to enforce a
patent against the party.
If you convey a covered work, knowingly relying on a patent license,
and the Corresponding Source of the work is not available for anyone
to copy, free of charge and under the terms of this License, through a
publicly available network server or other readily accessible means,
then you must either (1) cause the Corresponding Source to be so
available, or (2) arrange to deprive yourself of the benefit of the
patent license for this particular work, or (3) arrange, in a manner
consistent with the requirements of this License, to extend the patent
license to downstream recipients. "Knowingly relying" means you have
actual knowledge that, but for the patent license, your conveying the
covered work in a country, or your recipient's use of the covered work
in a country, would infringe one or more identifiable patents in that
country that you have reason to believe are valid.
If, pursuant to or in connection with a single transaction or
arrangement, you convey, or propagate by procuring conveyance of, a
covered work, and grant a patent license to some of the parties
receiving the covered work authorizing them to use, propagate, modify
or convey a specific copy of the covered work, then the patent license
you grant is automatically extended to all recipients of the covered
work and works based on it.
A patent license is "discriminatory" if it does not include within
the scope of its coverage, prohibits the exercise of, or is
conditioned on the non-exercise of one or more of the rights that are
specifically granted under this License. You may not convey a covered
work if you are a party to an arrangement with a third party that is
in the business of distributing software, under which you make payment
to the third party based on the extent of your activity of conveying
the work, and under which the third party grants, to any of the
parties who would receive the covered work from you, a discriminatory
patent license (a) in connection with copies of the covered work
conveyed by you (or copies made from those copies), or (b) primarily
for and in connection with specific products or compilations that
contain the covered work, unless you entered into that arrangement,
or that patent license was granted, prior to 28 March 2007.
Nothing in this License shall be construed as excluding or limiting
any implied license or other defenses to infringement that may
otherwise be available to you under applicable patent law.
12. No Surrender of Others' Freedom.
If conditions are imposed on you (whether by court order, agreement or
otherwise) that contradict the conditions of this License, they do not
excuse you from the conditions of this License. If you cannot convey a
covered work so as to satisfy simultaneously your obligations under this
License and any other pertinent obligations, then as a consequence you may
not convey it at all. For example, if you agree to terms that obligate you
to collect a royalty for further conveying from those to whom you convey
the Program, the only way you could satisfy both those terms and this
License would be to refrain entirely from conveying the Program.
13. Use with the GNU Affero General Public License.
Notwithstanding any other provision of this License, you have
permission to link or combine any covered work with a work licensed
under version 3 of the GNU Affero General Public License into a single
combined work, and to convey the resulting work. The terms of this
License will continue to apply to the part which is the covered work,
but the special requirements of the GNU Affero General Public License,
section 13, concerning interaction through a network will apply to the
combination as such.
14. Revised Versions of this License.
The Free Software Foundation may publish revised and/or new versions of
the GNU General Public License from time to time. Such new versions will
be similar in spirit to the present version, but may differ in detail to
address new problems or concerns.
Each version is given a distinguishing version number. If the
Program specifies that a certain numbered version of the GNU General
Public License "or any later version" applies to it, you have the
option of following the terms and conditions either of that numbered
version or of any later version published by the Free Software
Foundation. If the Program does not specify a version number of the
GNU General Public License, you may choose any version ever published
by the Free Software Foundation.
If the Program specifies that a proxy can decide which future
versions of the GNU General Public License can be used, that proxy's
public statement of acceptance of a version permanently authorizes you
to choose that version for the Program.
Later license versions may give you additional or different
permissions. However, no additional obligations are imposed on any
author or copyright holder as a result of your choosing to follow a
later version.
15. Disclaimer of Warranty.
THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
16. Limitation of Liability.
IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
SUCH DAMAGES.
17. Interpretation of Sections 15 and 16.
If the disclaimer of warranty and limitation of liability provided
above cannot be given local legal effect according to their terms,
reviewing courts shall apply local law that most closely approximates
an absolute waiver of all civil liability in connection with the
Program, unless a warranty or assumption of liability accompanies a
copy of the Program in return for a fee.
END OF TERMS AND CONDITIONS
How to Apply These Terms to Your New Programs
If you develop a new program, and you want it to be of the greatest
possible use to the public, the best way to achieve this is to make it
free software which everyone can redistribute and change under these terms.
To do so, attach the following notices to the program. It is safest
to attach them to the start of each source file to most effectively
state the exclusion of warranty; and each file should have at least
the "copyright" line and a pointer to where the full notice is found.
<one line to give the program's name and a brief idea of what it does.>
Copyright (C) <year> <name of author>
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.
Also add information on how to contact you by electronic and paper mail.
If the program does terminal interaction, make it output a short
notice like this when it starts in an interactive mode:
<program> Copyright (C) <year> <name of author>
This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
This is free software, and you are welcome to redistribute it
under certain conditions; type `show c' for details.
The hypothetical commands `show w' and `show c' should show the appropriate
parts of the General Public License. Of course, your program's commands
might be different; for a GUI interface, you would use an "about box".
You should also get your employer (if you work as a programmer) or school,
if any, to sign a "copyright disclaimer" for the program, if necessary.
For more information on this, and how to apply and follow the GNU GPL, see
<https://www.gnu.org/licenses/>.
The GNU General Public License does not permit incorporating your program
into proprietary programs. If your program is a subroutine library, you
may consider it more useful to permit linking proprietary applications with
the library. If this is what you want to do, use the GNU Lesser General
Public License instead of this License. But first, please read
<https://www.gnu.org/licenses/why-not-lgpl.html>.

118
README.md
Unescape View File

@ -1,2 +1,118 @@
# style-transfer-app
# 风格转换应用
该项目是一个风格转换应用，使用Vue.js和axios构建前端，后端使用Python的Flask框架。该应用允许用户上传两张图片，一张作为风格图片，另一张作为内容图片，并生成一张风格转换后的图片。
## 依赖安装
在运行该应用程序之前，请确保已安装以下软件和库：
- Python 3.x
- PyTorch （仅在1.11版本测试）
- torchvision
- Flask
- flask_cors
你可以使用pip命令安装它们：
```
pip install torch torchvision Flask flask_cors tqdm
```
安装前端依赖：
```bash
cd src/static
npm install
```
## 运行应用程序
要运行该应用程序，请按照以下步骤操作：
1. 在应用程序所在的目录中打开终端或命令提示符，确保已下载好相关u数据集及模型。
2. 运行以下命令启动应用程序：
```
python app.py
```
3. 应用程序将在本地主机上的默认端口（通常是http://127.0.0.1:5000/）上启动。
4. 安装前端依赖：
```bash
cd static
npm install
```
5. 启动前端应用：
```bash
npm run serve
```
## 使用方法
1. 打开浏览器并访问前端应用地址：`http://localhost:8080/`。
2. 页面会显示一个标题和两个图片上传框，分别对应风格图片和内容图片。
3. 点击第一个图片上传框，选择一个图片作为风格图片，支持JPEG和PNG格式。
4. 点击第二个图片上传框，选择一个图片作为内容图片，同样支持JPEG和PNG格式。
5. 在选择完两张图片后，点击“生成”按钮。
6. 等待一段时间，风格转换后的图片会在页面中显示出来。
## 注意事项
- 确保您的电脑上已经安装了Node.js和Python。
- 该应用仅支持JPEG和PNG格式的图片。
- 图片上传可能需要一定时间，取决于图片大小和网络速度。
- 风格转换后的图片会覆盖之前生成的结果，刷新页面将清除之前的结果。
## 技术实现
- 前端使用Vue.js框架构建，通过axios进行与后端的HTTP通信。
- 后端使用Python的Flask框架，接收前端上传的图片，进行风格转换并返回结果。
- 风格转换算法可能需要一定时间，请耐心等待。
## 下载链接
- [coco数据集](http://images.cocodataset.org/zips/train2017.zip)
- [wikiart数据集](https://www.kaggle.com/datasets/ipythonx/wikiart-gangogh-creating-art-gan)
- [基于vgg19训练的模型](https://pan.immiq.top/s/ZgSi9kY4yYiLWzb)
## 训练模型
1. 环境设置
要运行该代码，你需要满足以下环境要求：
- Python 3.6+
- PyTorch （仅在1.11下测试）
- torchvision
- numpy
- OpenCV
- tqdm
- matplotlib
- horovod
可以使用以下命令安装所需的依赖项：
```shell
pip install torch torchvision numpy opencv-python tqdm matplotlib horovod
```
2. 将内容图像和风格图像分别放置文件夹中，参考注释修改train.py中的数据集路径。
3. 运行以下命令开始训练模型：
```shell
python train.py
```
训练过程将持续多个周期（epochs），可以在代码中设置训练参数来控制训练的细节。
4. 训练完成后，模型将被保存在以下路径：
- MetaNet模型：'/root/autodl-tmp/improve/models/model_name.pth'
- TransformNet模型：'/root/autodl-tmp/improve/models/model_name_transform_net.pth'
请将上述路径中的'model_name'替换为你自己的模型名称。

183
src/app.py
Unescape View File

@ -0,0 +1,183 @@
from glob import glob
from tqdm import tqdm
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
import torchvision.models as models
from models import *
from utils import *
import subprocess
from flask import Flask, request, send_file, make_response
from flask_cors import CORS
app = Flask(__name__)
CORS(app) # 添加CORS支持
# 定义MetaNet类，继承自nn.Module
class MetaNet(nn.Module):
def __init__(self, param_dict):
super(MetaNet, self).__init__()
self.param_num = len(param_dict)
self.hidden = nn.Linear(1920, 128 * self.param_num)
self.fc_dict = {}
for i, (name, params) in enumerate(param_dict.items()):
self.fc_dict[name] = i
setattr(self, 'fc{}'.format(i + 1), nn.Linear(128, params))
def forward(self, mean_std_features):
hidden = F.relu(self.hidden(mean_std_features))
filters = {}
for name, i in self.fc_dict.items():
fc = getattr(self, 'fc{}'.format(i + 1))
filters[name] = fc(hidden[:, i * 128:(i + 1) * 128])
return list(filters.values())
def forward2(self, mean_std_features):
hidden = F.relu(self.hidden(mean_std_features))
filters = {}
for name, i in self.fc_dict.items():
fc = getattr(self, 'fc{}'.format(i + 1))
filters[name] = fc(hidden[:, i * 128:(i + 1) * 128])
return filters
width = 256
# 定义数据转换的操作
data_transform = transforms.Compose([
transforms.RandomResizedCrop(width, scale=(256 / 480, 1), ratio=(1, 1)),
transforms.ToTensor(),
tensor_normalizer
])
@app.route('/style-transfer', methods=['POST'])
def style_transfer():
# 获取上传的文件
style_image = request.files['style_image']
content_image = request.files['content_image']
# 保存上传的文件
style_path = 'style.jpg'
content_path = 'content.jpg'
style_image.save(style_path)
content_image.save(content_path)
# 读取style_image，并使用vgg_model提取其特征
style_image = read_image(
style_path,
target_width=256).to(device)
style_features = vgg_model(style_image)
style_mean_std = mean_std(style_features)
num_batches = 50
# 使用tqdm显示训练进度
with tqdm(enumerate(content_data_loader), total=num_batches) as pbar:
for batch, (content_images, _) in pbar:
x = content_images.cpu().numpy()
if (x.min(-1).min(-1) == x.max(-1).max(-1)).any():
continue
optimizer.zero_grad()
# 使用metanet获取权重
weights = metanet.forward2(mean_std(style_features))
transform_net.set_weights(weights, 0)
content_images = content_images.to(device)
transformed_images = transform_net(content_images)
content_features = vgg_model(content_images)
transformed_features = vgg_model(transformed_images)
transformed_mean_std = mean_std(transformed_features)
# 计算内容损失
content_loss = content_weight * \
F.mse_loss(transformed_features[2], content_features[2])
# 计算风格损失
style_loss = style_weight * \
F.mse_loss(transformed_mean_std, style_mean_std.expand_as(transformed_mean_std))
y = transformed_images
# 计算总变差损失
tv_loss = tv_weight * (torch.sum(torch.abs(y[:, :, :, :-1] - y[:, :, :, 1:])) + torch.sum(
torch.abs(y[:, :, :-1, :] - y[:, :, 1:, :])))
loss = content_loss + style_loss + tv_loss
loss.backward()
optimizer.step()
if batch > num_batches:
break
content_image = read_image(content_path)
content_image = tuple(content_image) # Convert the tensor to a tuple
content_image = torch.stack(content_image).to(device)
transformed_image = transform_net(content_image)
torchvision.utils.save_image(transformed_image, 'pics/result.jpg')
# 返回输出图片给前端
result_path = 'pics/result.jpg'
return send_file(result_path, mimetype='image/jpeg')
if __name__ == '__main__':
# 加载预训练的vgg19模型，并将其部分特征提取层赋值给vgg_model
vgg19_model = models.vgg19(pretrained=True)
vgg_model = VGG(vgg19_model.features[:36]).to(device).eval()
base = 32
# 创建一个TransformNet实例，base参数为32
transform_net = TransformNet(base).to(device)
# 获取transform_net的参数字典
transform_net.get_param_dict()
# 创建MetaNet实例，使用transform_net的参数字典作为输入
metanet = MetaNet(transform_net.get_param_dict()).to(device)
# 创建content_dataset数据集
content_dataset = torchvision.datasets.ImageFolder(
'/root/autodl-tmp/coco', transform=data_transform)
style_weight = 50
content_weight = 1
tv_weight = 1e-6
batch_size = 4
trainable_params = {}
trainable_param_shapes = {}
# 遍历vgg_model、transform_net和metanet的参数，将需要训练的参数添加到trainable_params字典中
for model in [vgg_model, transform_net, metanet]:
for name, param in model.named_parameters():
if param.requires_grad:
trainable_params[name] = param
trainable_param_shapes[name] = param.shape
# 使用Adam优化器，学习率为1e-3，对trainable_params中的参数进行优化
optimizer = optim.Adam(trainable_params.values(), 1e-3)
# 创建content_data_loader，用于加载content_dataset的数据
content_data_loader = torch.utils.data.DataLoader(
content_dataset, batch_size=batch_size, shuffle=True)
# 加载预训练的metanet和transform_net模型的权重
metanet.load_state_dict(torch.load(
'models-std/metanet_base32_style50_tv1e-06_tagnohvd.pth'))
transform_net.load_state_dict(torch.load(
'models-std/metanet_base32_style50_tv1e-06_tagnohvd_transform_net.pth'))
app.run()

32
src/copy_data.py
Unescape View File

@ -0,0 +1,32 @@
import os
import random
import shutil
def copy_images(dataset_dir, target_dir, percentage):
for root, dirs, files in os.walk(dataset_dir):
for file in files:
if random.random() <= percentage:
# 构建源文件路径和目标文件路径
source_path = os.path.join(root, file)
relative_path = os.path.relpath(source_path, dataset_dir)
target_path = os.path.join(target_dir, relative_path)
# 创建目标文件所在的文件夹
os.makedirs(os.path.dirname(target_path), exist_ok=True)
# 复制文件
shutil.copy2(source_path, target_path)
# 指定数据集目录和目标目录
dataset_dir_1 = '/root/autodl-tmp/wikiart/wikiart/images'
target_dir_1 = '/root/autodl-tmp/wikiart-small/wikiart/images'
dataset_dir_2 = '/root/autodl-tmp/coco'
target_dir_2 = '/root/autodl-tmp/coco-small'
# 指定要复制的图像百分比（1%）
percentage = 0.01
# 复制图像到目标目录
copy_images(dataset_dir_1, target_dir_1, percentage)
copy_images(dataset_dir_2, target_dir_2, percentage)

276
src/models.py
Unescape View File

@ -0,0 +1,276 @@
from collections import defaultdict
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.utils.data.distributed as distributed
import torchvision
import torchvision.transforms as transforms
import torchvision.models as models
from utils import *
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
class VGG(nn.Module):
def __init__(self, features):
"""
初始化VGG模型
Args:
features (nn.Module): VGG模型的特征提取部分
"""
super(VGG, self).__init__()
self.features = features
self.layer_name_mapping = {
'3': "relu1_2",
'8': "relu2_2",
'17': "relu3_3",
'26': "relu4_3"
} # 定义VGG模型中各层的名称映射
for p in self.parameters():
p.requires_grad = False # 冻结模型的参数，使其不可训练
def forward(self, x):
"""
VGG模型的前向传播
Args:
x (torch.Tensor): 输入张量
Returns:
List[torch.Tensor]: 指定层的输出列表
"""
outs = []
for name, module in self.features._modules.items():
x = module(x)
if name in self.layer_name_mapping:
outs.append(x) # 提取指定层的输出
return outs
class MyConv2D(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size=3, stride=1):
"""
初始化自定义的二维卷积层
Args:
in_channels (int): 输入通道数
out_channels (int): 输出通道数
kernel_size (int or tuple): 卷积核大小，默认为3
stride (int or tuple): 步长，默认为1
"""
super(MyConv2D, self).__init__()
self.weight = torch.zeros(
(out_channels,
in_channels,
kernel_size,
kernel_size)).to(device) # 创建卷积核权重
self.bias = torch.zeros(out_channels).to(device) # 创建卷积核偏置
self.in_channels = in_channels # 输入通道数
self.out_channels = out_channels # 输出通道数
self.kernel_size = (kernel_size, kernel_size) # 卷积核大小
self.stride = (stride, stride) # 步长
def forward(self, x):
return F.conv2d(x, self.weight, self.bias, self.stride) # 执行卷积运算
def extra_repr(self):
"""
返回额外的模块信息字符串
Returns:
str: 模块信息字符串
"""
s = ('{in_channels}, {out_channels}, kernel_size={kernel_size}'
', stride={stride}') # 打印额外的模块信息
return s.format(**self.__dict__)
class ResidualBlock(nn.Module):
def __init__(self, channels):
"""
初始化残差块
Args:
channels (int): 输入和输出通道数
"""
super(ResidualBlock, self).__init__()
self.conv = nn.Sequential(
*ConvLayer(channels, channels, kernel_size=3, stride=1),
*ConvLayer(channels, channels, kernel_size=3, stride=1, relu=False)
) # 定义残差块的卷积层序列
def forward(self, x):
"""
执行残差块的前向传播
Args:
x (tensor): 输入张量
Returns:
tensor: 输出张量
"""
return self.conv(x) + x # 执行残差块的前向传播，加上输入张量
def ConvLayer(in_channels, out_channels, kernel_size=3, stride=1,
upsample=None, instance_norm=True, relu=True, trainable=False):
"""
构建卷积层的序列
Args:
in_channels (int): 输入通道数
out_channels (int): 输出通道数
kernel_size (int): 卷积核大小
stride (int): 步长
upsample (int or None): 上采样因子，默认为None
instance_norm (bool): 是否使用实例归一化层，默认为True
relu (bool): 是否使用ReLU激活函数，默认为True
trainable (bool): 是否使用可训练的卷积层，默认为False
Returns:
list: 卷积层序列
"""
layers = []
if upsample:
layers.append(
nn.Upsample(
mode='nearest',
scale_factor=upsample)) # 上采样操作
layers.append(nn.ReflectionPad2d(kernel_size // 2)) # 反射填充
if trainable:
layers.append(
nn.Conv2d(
in_channels,
out_channels,
kernel_size,
stride)) # 可训练的卷积层
else:
layers.append(
MyConv2D(
in_channels,
out_channels,
kernel_size,
stride)) # 自定义的卷积层
if instance_norm:
layers.append(nn.InstanceNorm2d(out_channels)) # 实例归一化层
if relu:
layers.append(nn.ReLU()) # ReLU激活函数
return layers # 返回卷积层序列
class TransformNet(nn.Module):
def __init__(self, base=8):
super(TransformNet, self).__init__()
self.base = base
self.weights = [] # 权重列表
self.downsampling = nn.Sequential(
*ConvLayer(3, base, kernel_size=9, trainable=True), # 下采样层
*ConvLayer(base, base * 2, kernel_size=3, stride=2),
*ConvLayer(base * 2, base * 4, kernel_size=3, stride=2),
)
self.residuals = nn.Sequential(
*[ResidualBlock(base * 4) for i in range(5)]) # 残差块
self.upsampling = nn.Sequential(
*
ConvLayer(
base *
4,
base *
2,
kernel_size=3,
upsample=2), # 上采样层
*
ConvLayer(
base *
2,
base,
kernel_size=3,
upsample=2),
*
ConvLayer(
base,
3,
kernel_size=9,
instance_norm=False,
relu=False,
trainable=True),
)
self.get_param_dict()
def forward(self, X):
y = self.downsampling(X) # 下采样
y = self.residuals(y) # 残差块
y = self.upsampling(y) # 上采样
return y
def get_param_dict(self):
param_dict = defaultdict(int) # 参数字典
def dfs(module, name):
for name2, layer in module.named_children():
dfs(layer, '%s.%s' % (name, name2) if name != '' else name2)
if module.__class__ == MyConv2D:
param_dict[name] += int(np.prod(module.weight.shape))
param_dict[name] += int(np.prod(module.bias.shape))
dfs(self, '')
return param_dict
def set_my_attr(self, name, value):
target = self
for x in name.split('.'):
if x.isnumeric():
target = target.__getitem__(int(x))
else:
target = getattr(target, x)
n_weight = np.prod(target.weight.shape)
target.weight = value[:n_weight].view(target.weight.shape) # 设置权重
target.bias = value[n_weight:].view(target.bias.shape) # 设置偏置项
def set_weights(self, weights, i=0):
for name, param in weights.items():
self.set_my_attr(name, weights[name][i]) # 设置权重
class MetaNet(nn.Module):
def __init__(self, param_dict):
"""
MetaNet模型的构造函数
Args:
param_dict (dict): 参数字典，包含线性层名称和对应的输出维度
"""
super(MetaNet, self).__init__()
self.param_num = len(param_dict) # 参数数量
self.hidden = nn.Linear(1920, 128 * self.param_num) # 隐藏层
self.fc_dict = {}
for i, (name, params) in enumerate(param_dict.items()):
self.fc_dict[name] = i
setattr(self, 'fc{}'.format(i + 1), nn.Linear(128, params))
def forward(self, mean_std_features):
"""
MetaNet模型的前向传播
Args:
mean_std_features (torch.Tensor): 均值和标准差特征向量，形状为 (batch_size, 1920)
Returns:
dict: 线性层输出字典，包含线性层名称和对应的输出张量
"""
hidden = F.relu(self.hidden(mean_std_features)) # 隐藏层
filters = {}
for name, i in self.fc_dict.items():
fc = getattr(self, 'fc{}'.format(i + 1))
filters[name] = fc(hidden[:, i * 128:(i + 1) * 128]) # 线性层
return filters

23
src/static/.gitignore vendored
Unescape View File

@ -0,0 +1,23 @@
.DS_Store
node_modules
/dist
# local env files
.env.local
.env.*.local
# Log files
npm-debug.log*
yarn-debug.log*
yarn-error.log*
pnpm-debug.log*
# Editor directories and files
.idea
.vscode
*.suo
*.ntvs*
*.njsproj
*.sln
*.sw?

23
src/static/README.md
Unescape View File

@ -0,0 +1,23 @@
# numb
## Project setup
```
npm install
```
### Compiles and hot-reloads for development
```
npm run serve
```
### Compiles and minifies for production
```
npm run build
```
### Customize configuration
See [Configuration Reference](https://cli.vuejs.org/config/).

5
src/static/babel.config.js
Unescape View File

@ -0,0 +1,5 @@
module.exports = {
presets: [
'@vue/cli-plugin-babel/preset'
]
}

19
src/static/jsconfig.json
Unescape View File

@ -0,0 +1,19 @@
{
"compilerOptions": {
"target": "es5",
"module": "esnext",
"baseUrl": "./",
"moduleResolution": "node",
"paths": {
"@/*": [
"src/*"
]
},
"lib": [
"esnext",
"dom",
"dom.iterable",
"scripthost"
]
}
}

10093
src/static/package-lock.json generated
View File

File diff suppressed because it is too large Load Diff

32
src/static/package.json
Unescape View File

@ -0,0 +1,32 @@
{
"name": "numb",
"version": "0.1.0",
"private": true,
"scripts": {
"serve": "vue-cli-service serve",
"build": "vue-cli-service build"
},
"dependencies": {
"axios": "^1.4.0",
"core-js": "^3.8.3",
"element-ui": "^2.15.13",
"vue": "^2.6.14",
"vue-axios": "^3.5.2",
"vue-router": "^3.5.1",
"vuex": "^3.6.2"
},
"devDependencies": {
"@vue/cli-plugin-babel": "~5.0.0",
"@vue/cli-plugin-router": "~5.0.0",
"@vue/cli-plugin-vuex": "~5.0.0",
"@vue/cli-service": "~5.0.0",
"less": "^4.0.0",
"less-loader": "^8.0.0",
"vue-template-compiler": "^2.6.14"
},
"browserslist": [
"> 1%",
"last 2 versions",
"not dead"
]
}

BIN
src/static/public/favicon.ico
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 4.2 KiB

18
src/static/public/index.html
Unescape View File

@ -0,0 +1,18 @@
<!DOCTYPE html>
<html lang="">
<head>
<meta charset="utf-8">
<meta content="IE=edge" http-equiv="X-UA-Compatible">
<meta content="width=device-width,initial-scale=1.0" name="viewport">
<link href="<%= BASE_URL %>favicon.ico" rel="icon">
<title><%= htmlWebpackPlugin.options.title %></title>
</head>
<body>
<noscript>
<strong>We're sorry but <%= htmlWebpackPlugin.options.title %> doesn't work properly without JavaScript enabled.
Please enable it to continue.</strong>
</noscript>
<div id="app"></div>
<!-- built files will be auto injected -->
</body>
</html>

29
src/static/src/App.vue
Unescape View File

@ -0,0 +1,29 @@
<template>
<div id="app">
<router-view/>
</div>
</template>
<style lang="less">
#app {
font-family: Avenir, Helvetica, Arial, sans-serif;
-webkit-font-smoothing: antialiased;
-moz-osx-font-smoothing: grayscale;
text-align: center;
color: #2c3e50;
}
nav {
padding: 30px;
a {
font-weight: bold;
color: #2c3e50;
&.router-link-exact-active {
color: #42b983;
}
}
}
</style>

BIN
src/static/src/assets/logo.png
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 6.7 KiB

66
src/static/src/components/HelloWorld.vue
Unescape View File

@ -0,0 +1,66 @@
<template>
<div class="hello">
<h1>{{ msg }}</h1>
<p>
For a guide and recipes on how to configure / customize this project,<br>
check out the
<a href="https://cli.vuejs.org" rel="noopener" target="_blank">vue-cli documentation</a>.
</p>
<h3>Installed CLI Plugins</h3>
<ul>
<li><a href="https://github.com/vuejs/vue-cli/tree/dev/packages/%40vue/cli-plugin-babel" rel="noopener"
target="_blank">babel</a></li>
<li><a href="https://github.com/vuejs/vue-cli/tree/dev/packages/%40vue/cli-plugin-router" rel="noopener"
target="_blank">router</a></li>
<li><a href="https://github.com/vuejs/vue-cli/tree/dev/packages/%40vue/cli-plugin-vuex" rel="noopener"
target="_blank">vuex</a></li>
</ul>
<h3>Essential Links</h3>
<ul>
<li><a href="https://vuejs.org" rel="noopener" target="_blank">Core Docs</a></li>
<li><a href="https://forum.vuejs.org" rel="noopener" target="_blank">Forum</a></li>
<li><a href="https://chat.vuejs.org" rel="noopener" target="_blank">Community Chat</a></li>
<li><a href="https://twitter.com/vuejs" rel="noopener" target="_blank">Twitter</a></li>
<li><a href="https://news.vuejs.org" rel="noopener" target="_blank">News</a></li>
</ul>
<h3>Ecosystem</h3>
<ul>
<li><a href="https://router.vuejs.org" rel="noopener" target="_blank">vue-router</a></li>
<li><a href="https://vuex.vuejs.org" rel="noopener" target="_blank">vuex</a></li>
<li><a href="https://github.com/vuejs/vue-devtools#vue-devtools" rel="noopener" target="_blank">vue-devtools</a>
</li>
<li><a href="https://vue-loader.vuejs.org" rel="noopener" target="_blank">vue-loader</a></li>
<li><a href="https://github.com/vuejs/awesome-vue" rel="noopener" target="_blank">awesome-vue</a></li>
</ul>
</div>
</template>
<script>
export default {
name: 'HelloWorld',
props: {
msg: String
}
}
</script>
<!-- Add "scoped" attribute to limit CSS to this component only -->
<style lang="less" scoped>
h3 {
margin: 40px 0 0;
}
ul {
list-style-type: none;
padding: 0;
}
li {
display: inline-block;
margin: 0 10px;
}
a {
color: #42b983;
}
</style>

19
src/static/src/main.js
Unescape View File

@ -0,0 +1,19 @@
import Vue from 'vue'
import App from './App.vue'
import router from './router'
import store from './store'
import axios from 'axios'
import VueAxios from 'vue-axios'
import ElementUI from 'element-ui';
import 'element-ui/lib/theme-chalk/index.css';
Vue.config.productionTip = false
Vue.use(ElementUI);
Vue.use(VueAxios, axios)
new Vue({
router,
store,
render: h => h(App)
}).$mount('#app')

27
src/static/src/router/index.js
Unescape View File

@ -0,0 +1,27 @@
import Vue from 'vue'
import VueRouter from 'vue-router'
import Home from '../views/Home.vue'
Vue.use(VueRouter)
const routes = [
{
path: '/',
name: 'Home',
component: Home
},
{
path: '/About',
name: 'About',
// route level code-splitting
// this generates a separate chunk (about.[hash].js) for this route
// which is lazy-loaded when the route is visited.
component: () => import(/* webpackChunkName: "about" */ '../views/About.vue')
}
]
const router = new VueRouter({
routes
})
export default router

12
src/static/src/store/index.js
Unescape View File

@ -0,0 +1,12 @@
import Vue from 'vue'
import Vuex from 'vuex'
Vue.use(Vuex)
export default new Vuex.Store({
state: {},
getters: {},
mutations: {},
actions: {},
modules: {}
})

0
src/static/src/views/About.vue
Unescape View File

100
src/static/src/views/Home.vue
Unescape View File

@ -0,0 +1,100 @@
<template>
<body>
<div>
<h1>风格转换</h1>
<div id="show-picture"></div>
<input id="user-photo" accept="image/jpeg, image/png" type="file" @change="handleStyleImageUpload "/>
<el-divider></el-divider>
<div id="show-picture2"></div>
<input accept="image/jpeg, image/png" type="file" @change="handleContentImageUpload"/>
<el-divider></el-divider>
<button :disabled="!styleImage || !contentImage" @click="generateStyledImage">生成</button>
<div v-if="styledImage">
<h2>转换后的图片</h2>
<img :src="styledImage" alt="Styled Image"/>
</div>
</div>
</body>
</template>
<script>
import axios from "axios";
export default {
name: 'Home',
data() {
return {
styleImage: null,
contentImage: null,
styledImage: null,
};
},
methods: {
handleStyleImageUpload(event) {
this.styleImage = event.target.files[0];
var reader = new FileReader();
reader.readAsDataURL(this.styleImage);
reader.onload = function () {
var image = document.createElement("img");
image.width = "400";
image.src = reader.result;
var showPicture = document.getElementById("show-picture");
showPicture.append(image);
}
},
handleContentImageUpload(event) {
this.contentImage = event.target.files[0];
var reader = new FileReader();
reader.readAsDataURL(this.contentImage);
reader.onload = function () {
var image = document.createElement("img");
image.width = "400";
image.src = reader.result;
var showPicture = document.getElementById("show-picture2");
showPicture.append(image);
}
},
generateStyledImage() {
const formData = new FormData();
formData.append('style_image', this.styleImage);
formData.append('content_image', this.contentImage);
axios.post('http://127.0.0.1:5000/style-transfer', formData, {responseType: 'blob'})
.then(response => {
const styledImage = URL.createObjectURL(response.data)
this.styledImage = styledImage;
this.$refs.resultImage.onload = () => {
URL.revokeObjectURL(styledImage);
};
})
.catch(error => {
console.error('Error generating styled image:', error);
});
},
},
}
</script>
<style>
body {
background: lightblue;
}
.pic1 {
position: page;
top: 100px;
left: 50px;
}
.pic2 {
position: fixed;
top: 40px;
right: 50px;
}
</style>

4
src/static/vue.config.js
Unescape View File

@ -0,0 +1,4 @@
const {defineConfig} = require('@vue/cli-service')
module.exports = defineConfig({
transpileDependencies: true
})

313
src/train.py
Unescape View File

@ -0,0 +1,313 @@
import random
from glob import glob
import multiprocessing
import pathlib
from collections import defaultdict
import numpy as np
import cv2
from tqdm import tqdm
from pprint import pprint
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.utils.data.distributed as distributed
import torchvision
import torchvision.transforms as transforms
import torchvision.models as models
from PIL import Image
import matplotlib.pyplot as plt
import horovod.torch as hvd
from models import *
from utils import *
display = pprint # 定义pprint别名为display
hvd.init() # 初始化Horovod
torch.cuda.set_device(hvd.local_rank()) # 设置当前设备为Horovod本地排名对应的GPU设备
device = torch.device(
"cuda:%s" %
hvd.local_rank() if torch.cuda.is_available() else "cpu") # 根据是否有可用的GPU设备选择设备类型
torch.multiprocessing.set_sharing_strategy(
'file_system') # 设置PyTorch多进程共享策略为文件系统
is_hvd = False # 是否使用Horovod进行分布式训练
tag = 'nohvd' # 标签
base = 32 # TransformNet的基础通道数
style_weight = 50 # 风格损失的权重
content_weight = 1 # 内容损失的权重
tv_weight = 1e-6 # 总变差损失的权重
epochs = 22 # 训练的轮数
batch_size = 32 # 批处理大小
width = 256 # 图像的宽度
verbose_hist_batch = 100 # 打印训练过程中的历史损失的批次间隔
verbose_image_batch = 800 # 打印生成图像的批次间隔
# 模型名称
model_name = 'metanet_base{}_style{}_tv{}_tag{}'.format(
base, style_weight, tv_weight, tag)
rank = hvd.rank() if hasattr(hvd, 'rank') else None # 获取当前进程的排名
# 打印模型名称和当前进程的排名
print('model_name: {}, rank: {}'.format(model_name, rank))
def remove_directory(path):
# 删除目录及其内容
try:
path = pathlib.Path(path)
path.unlink(missing_ok=True) # 删除文件或空目录
if path.is_dir():
for child in path.iterdir():
remove_directory(child) # 递归删除子目录或文件
path.rmdir() # 删除目录
except Exception as e:
# 处理异常情况
print(f"Failed to remove {path}: {e}")
# 删除隐藏文件
hidden_files = pathlib.Path('runs').glob('*/.AppleDouble')
for file in hidden_files:
remove_directory(file)
# 删除模型保存目录
model_path = pathlib.Path('runs') / model_name
remove_directory(model_path)
# 创建预训练的VGG19模型
vgg19 = models.vgg19(pretrained=True)
vgg_features = vgg19.features[:36] # 选择要提取特征的VGG层次
vgg = VGG(vgg_features).to(device).eval() # 构建截断的VGG网络，用于提取特征
# 创建TransformNet模型
transform_net = TransformNet(base).to(device)
transform_net_param_dict = transform_net.get_param_dict() # 获取TransformNet模型的参数字典
# 创建MetaNet模型
metanet = MetaNet(transform_net_param_dict).to(device)
# 数据预处理的转换操作
data_transform = transforms.Compose([
transforms.RandomResizedCrop(width, scale=(
256 / 480, 1), ratio=(1, 1)), # 随机裁剪图像
transforms.ToTensor(), # 转换为张量
tensor_normalizer # 标准化
])
style_dataset = torchvision.datasets.ImageFolder(
'/root/autodl-tmp/wikiart/wikiart/images', # 风格图像数据集路径
transform=data_transform) # 数据集的变换操作
content_dataset = torchvision.datasets.ImageFolder(
'/root/autodl-tmp/coco', transform=data_transform) # 内容图像数据集路径
if is_hvd:
# 创建分布式训练的采样器
train_sampler = distributed.DistributedSampler(
content_dataset, num_replicas=hvd.size(), rank=hvd.rank())
# 创建数据加载器，用于加载内容数据集
content_data_loader = torch.utils.data.DataLoader(
content_dataset,
batch_size=batch_size,
num_workers=multiprocessing.cpu_count(),
sampler=train_sampler)
else:
# 创建数据加载器，用于加载内容数据集
content_data_loader = torch.utils.data.DataLoader(
content_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=multiprocessing.cpu_count())
if not is_hvd or hvd.rank() == 0:
# 打印风格数据集信息和内容数据集信息
print(style_dataset)
print('-' * 20)
print(content_dataset)
# 设置模型为评估模式
metanet.eval()
transform_net.eval()
# 创建随机张量
rands = torch.rand(4, 3, 256, 256).to(device)
# 提取随机张量的特征
features = vgg(rands)
# 使用MetaNet计算特征的权重
weights = metanet(mean_std(features))
# 将权重设置到TransformNet模型中
transform_net.set_weights(weights)
# 使用TransformNet对随机张量进行变换
with torch.no_grad():
transformed_images = transform_net(torch.rand(4, 3, 256, 256).to(device))
# 打印特征的形状
if not is_hvd or hvd.rank() == 0:
print('Features:')
display([x.shape for x in features])
# 打印权重的形状
print('Weights:')
display([x.shape for x in weights.values()])
# 打印变换后图像的形状
print('Transformed Images:')
display(transformed_images.shape)
visualization_style_image = random.choice(
style_dataset)[0].unsqueeze(0).to(device) # 随机选择一个风格图像进行可视化
visualization_content_images = torch.stack([random.choice(
content_dataset)[0] for i in range(4)]).to(device) # 随机选择4个内容图像进行可视化
if not is_hvd or hvd.rank() == 0:
for f in glob('runs/*/.AppleDouble'):
remove_directory(f) # 删除隐藏文件
remove_directory('runs/' + model_name) # 删除模型保存目录
visualization_style_image = random.choice(
style_dataset)[0].unsqueeze(0).to(device) # 随机选择一个风格图像，并将其转换为张量，并移到指定设备上
visualization_content_images = torch.stack(
[random.choice(content_dataset)[0] for i in range(4)]).to(device) # 随机选择4个内容图像，并将它们转换为张量，并移到指定设备上
del rands, features, weights, transformed_images # 删除不再需要的变量，释放内存
trainable_params = {} # 可训练参数的字典
trainable_param_shapes = {} # 可训练参数的形状字典
for model in [vgg, transform_net, metanet]:
for name, param in model.named_parameters():
if param.requires_grad:
trainable_params[name] = param # 将可训练参数添加到字典中
trainable_param_shapes[name] = param.shape # 记录可训练参数的形状
# 创建Adam优化器，并传入可训练参数的值列表和学习率
optimizer = optim.Adam(trainable_params.values(), lr=1e-3)
if is_hvd:
# 使用Horovod分布式优化器
optimizer = hvd.DistributedOptimizer(
optimizer, named_parameters=trainable_params.items())
# 合并TransformNet和MetaNet的状态字典
params = transform_net.state_dict()
params.update(metanet.state_dict())
# 使用Horovod广播模型参数
hvd.broadcast_parameters(params, root_rank=0)
# 内容数据加载器的批次数
n_batch = len(content_data_loader)
# 设置MetaNet和TransformNet为训练模式
metanet.train()
transform_net.train()
for epoch in range(epochs):
smoother = defaultdict(Smooth) # 平滑器，用于计算平均损失
with tqdm(enumerate(content_data_loader), total=n_batch) as pbar: # 使用tqdm显示进度条
for batch, (content_images, _) in pbar:
n_iter = epoch * n_batch + batch # 当前迭代次数
if batch % 20 == 0:
style_image = random.choice(style_dataset)[0].unsqueeze(
0).to(device) # 每20个批次随机选择一个风格图像，并将其转换为张量，并移到指定设备上
style_features = vgg(style_image) # 提取风格图像的特征
style_mean_std = mean_std(style_features) # 计算风格特征的均值和标准差
x = content_images.cpu().numpy() # 将内容图像转移到CPU并转换为NumPy数组
if (x.min(-1).min(-1) == x.max(-1).max(-1)
).any(): # 如果存在像素值全相等的图像，则跳过该批次
continue
optimizer.zero_grad() # 梯度清零
weights = metanet(mean_std(style_features)) # 使用meta网络计算权重
transform_net.set_weights(weights, 0) # 将权重设置到转换网络中的第一个模块
content_images = content_images.to(device) # 将内容图像移动到指定设备上
transformed_images = transform_net(
content_images) # 使用转换网络对内容图像进行转换
content_features = vgg(content_images) # 提取内容图像的特征
transformed_features = vgg(transformed_images) # 提取转换后图像的特征
transformed_mean_std = mean_std(
transformed_features) # 计算转换后特征的均值和标准差
content_loss = content_weight * \
F.mse_loss(
transformed_features[2],
content_features[2]) # 计算内容损失
style_loss = style_weight * \
F.mse_loss(
transformed_mean_std,
style_mean_std.expand_as(transformed_mean_std)) # 计算风格损失
y = transformed_images
tv_loss = tv_weight * (torch.sum(torch.abs(y[:, :, :, :-1] - y[:, :, :, 1:])) +
torch.sum(torch.abs(y[:, :, :-1, :] - y[:, :, 1:, :]))) # 计算总变差损失
loss = content_loss + style_loss + tv_loss # 总损失
loss.backward() # 反向传播计算梯度
optimizer.step() # 更新模型参数
smoother['content_loss'] += content_loss.item() # 累加内容损失
smoother['style_loss'] += style_loss.item() # 累加风格损失
smoother['tv_loss'] += tv_loss.item() # 累加总变差损失
smoother['loss'] += loss.item() # 累加总损失
max_value = max([x.max().item()
for x in weights.values()]) # 计算权重的最大值
s = 'Epoch: {} '.format(epoch + 1) # 当前的epoch
# 平滑后的内容损失
s += 'Content: {:.2f} '.format(smoother['content_loss'])
s += 'Style: {:.1f} '.format(smoother['style_loss']) # 平滑后的风格损失
s += 'Loss: {:.2f} '.format(smoother['loss']) # 平滑后的总损失
s += 'Max: {:.2f}'.format(max_value) # 权重的最大值
if (batch + 1) % verbose_image_batch == 0:
transform_net.eval() # 设置转换网络为评估模式
visualization_transformed_images = transform_net(
visualization_content_images) # 使用转换网络对可视化内容图像进行转换
transform_net.train() # 设置转换网络为训练模式
visualization_transformed_images = torch.cat(
[style_image, visualization_transformed_images]) # 将风格图像与可视化转换图像拼接起来
del visualization_transformed_images # 删除变量，释放内存
pbar.set_description(s) # 更新进度条描述
del transformed_images, weights # 删除变量，释放内存
if not is_hvd or hvd.rank() == 0:
torch.save(metanet.state_dict(),
'/root/autodl-tmp/improve/checkpoints/{}_{}.pth'.format(model_name,
epoch + 1)) # 保存meta网络的模型参数
torch.save(
transform_net.state_dict(),
'/root/autodl-tmp/improve/checkpoints/{}_transform_net_{}.pth'.format(
model_name,
epoch + 1)) # 保存转换网络的模型参数
torch.save(
metanet.state_dict(),
'/root/autodl-tmp/improve/models/{}.pth'.format(model_name)) # 保存meta网络的模型参数
torch.save(
transform_net.state_dict(),
'/root/autodl-tmp/improve/models/{}_transform_net.pth'.format(model_name)) # 保存转换网络的模型参数

259
src/utils.py
Unescape View File

@ -0,0 +1,259 @@
import numpy as np
import cv2
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.utils.data.distributed as distributed
import torchvision
import torchvision.transforms as transforms
import torchvision.models as models
from PIL import Image
import matplotlib.pyplot as plt
cnn_normalization_mean = [0.485, 0.456, 0.406] # 图像归一化的均值
cnn_normalization_std = [0.229, 0.224, 0.225] # 图像归一化的标准差
tensor_normalizer = transforms.Normalize(
mean=cnn_normalization_mean,
std=cnn_normalization_std) # 创建图像张量的标准化转换器
epsilon = 1e-5 # 避免除以零的小量
def preprocess_image(
image,
target_width=None,
resize=True,
center_crop=True,
normalize=True):
"""
预处理图像，包括调整大小、剪裁、转换为张量并进行标准化
Args:
image (PIL.Image.Image): 输入的图像
target_width (int): 目标宽度 (default: None)
resize (bool): 是否调整大小 (default: True)
center_crop (bool): 是否居中裁剪 (default: True)
normalize (bool): 是否标准化 (default: True)
Returns:
torch.Tensor or None: 预处理后的图像张量，如果发生异常则返回 None
Raises:
RuntimeError: 当图像尺寸过大导致内存溢出时
"""
transforms_list = []
if resize and target_width:
transforms_list.append(transforms.Resize(target_width)) # 调整图像大小
if center_crop and target_width:
transforms_list.append(transforms.CenterCrop(target_width)) # 居中裁剪
transforms_list.append(transforms.ToTensor()) # 转换为张量
if normalize:
transforms_list.append(tensor_normalizer) # 标准化
transform = transforms.Compose(transforms_list)
try:
return transform(image).unsqueeze(0) # 添加一维作为批处理维度
except RuntimeError:
# 处理图像尺寸过大导致的内存溢出异常
print("Error: Image size too large.")
return None
def image_to_tensor(image, target_width=None):
"""
将图像转换为张量
Args:
image (numpy.ndarray): 输入的图像数组
target_width (int): 目标宽度 (default: None)
Returns:
torch.Tensor or None: 预处理后的图像张量，如果发生异常则返回 None
Raises:
ValueError: 当输入图像为无效类型或尺寸时
"""
try:
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) # 转换颜色通道顺序为RGB
image = Image.fromarray(image) # 将数组转换为图像
return preprocess_image(image, target_width) # 预处理图像
except ValueError as e:
print(f"Error: Invalid input image. {str(e)}")
return None
def read_image(path, target_width=None):
"""
从文件路径读取图像并进行预处理
Args:
path (str): 图像文件的路径
target_width (int): 目标宽度 (default: None)
Returns:
torch.Tensor or None: 预处理后的图像张量，如果发生异常则返回 None
Raises:
FileNotFoundError: 当文件路径指定的图像文件不存在
ValueError: 当输入图像尺寸无效或预处理过程中发生异常
"""
try:
image = Image.open(path) # 打开图像文件
return preprocess_image(image, target_width) # 预处理图像
except FileNotFoundError:
print(f"Error: Image file not found at path: {path}")
return None
except ValueError as e:
print(f"Error: Invalid input image. {str(e)}")
return None
def recover_image(tensor):
"""
将张量转换回图像
Args:
tensor (torch.Tensor): 输入的张量
Returns:
numpy.ndarray: 恢复后的图像数组
"""
image = tensor.to(torch.uint8).detach(
).cpu().numpy() # 将张量转移到CPU并转换为NumPy数组
image = image * np.array(cnn_normalization_std).reshape((1, 3, 1, 1)) + \
np.array(cnn_normalization_mean).reshape((1, 3, 1, 1))
return (
image.transpose(
0,
2,
3,
1) *
255.).clip(
0,
255).astype(
np.uint8)[0]
def recover_tensor(tensor):
"""
将张量进行反标准化
Args:
tensor (torch.Tensor): 输入的张量
Returns:
torch.Tensor: 反标准化后的张量，数值在0和1之间
"""
mean = torch.tensor(cnn_normalization_mean).reshape(
(1, 3, 1, 1)).to(
tensor.device)
std = torch.tensor(cnn_normalization_std).reshape(
(1, 3, 1, 1)).to(
tensor.device)
tensor = tensor * std + mean # 反标准化
tensor = torch.clamp(tensor, 0, 1) # 对张量进行裁剪，确保数值在0和1之间
return tensor
def imshow(tensor, title=None):
"""
显示图像张量
Args:
tensor (torch.Tensor): 输入的张量
title (str or None): 图像的标题 (default: None)
"""
image = recover_image(tensor) # 将张量转换为图像
image = image.astype(np.uint8) # 将图像数组转换为无符号整数类型
image = np.clip(image, 0, 255) # 对图像进行裁剪，确保数值在0和255之间
image = image[..., ::-1] # 调整图像的颜色通道顺序为RGB
plt.imshow(image) # 显示图像
if title is not None:
plt.title(title) # 设置标题
plt.show() # 显示图像
def mean_std(features, epsilon=1e-5):
"""
计算特征的均值和标准差
Args:
features (List[torch.Tensor]): 特征列表
epsilon (float): 避免除以零的小值 (default: 1e-5)
Returns:
torch.Tensor: 均值和标准差的张量
"""
mean_std_features = []
for x in features:
x = x.view(*x.shape[:2], -1) # 重塑形状以便计算均值和方差
x = torch.cat([x.mean(-1),
torch.sqrt(x.var(-1) + epsilon)],
dim=-1) # 计算均值和标准差
n = x.shape[0]
# 【mean, ..., std, ...] to [mean, std, ...]
x2 = x.view(n, 2, -1).transpose(2,
1).contiguous().view(n, -1) # 重排张量的顺序
mean_std_features.append(x2) # 添加到列表中
mean_std_features = torch.cat(mean_std_features, dim=-1) # 按维度拼接张量
return mean_std_features # 返回均值和标准差的张量
class Smooth:
def __init__(self, window_size=100):
"""
初始化Smooth类
Args:
window_size (int): 平滑窗口的大小 (default: 100)
"""
self.window_size = window_size # 平滑窗口的大小
self.data = np.zeros(
(self.window_size, 1), dtype=np.float32) # 存储数据的数组
self.index = 0 # 当前索引
def __iadd__(self, x):
"""
实现+=运算符重载
Args:
x (float): 要添加到数据数组中的值
Returns:
Smooth: 更新后的Smooth对象
"""
if self.index == 0:
self.data[:] = x
self.data[self.index % self.window_size] = x # 更新数据数组中的值
self.index += 1 # 增加索引
return self
def __float__(self):
"""
实现float()函数转换
Returns:
float: 数据的平均值
"""
return float(self.data.mean()) # 返回数据的平均值
def __format__(self, f):
"""
实现格式化输出
Args:
f (str): 格式字符串
Returns:
str: 格式化后的平均值字符串
"""
return self.__float__().__format__(f) # 格式化输出平均值
Write Preview
Loading…
Cancel
Save