Skip to content

GitLab

Commit fccfdfa5 authored by dlyrm's avatar dlyrm
Browse files

update code

parent dcc7bf4f
Pipeline #681 canceled with stages
Hide whitespace changes
Inline Side-by-side
Showing with 2271 additions and 1 deletion
.gitignore 0 → 100644
View file @ fccfdfa5
# Virtualenv
/.venv/
/venv/
# Byte-compiled / optimized / DLL files
__pycache__/
.ipynb_checkpoints/
*.py[cod]
# C extensions
*.so
# json file
*.json
# log file
*.log
# Distribution / packaging
/bin/
/build/
/develop-eggs/
/dist/
/eggs/
/lib/
/lib64/
/output/
/inference_model/
/output_inference/
/parts/
/sdist/
/var/
/*.egg-info/
/.installed.cfg
/*.egg
/.eggs
# AUTHORS and ChangeLog will be generated while packaging
/AUTHORS
/ChangeLog
# BCloud / BuildSubmitter
/build_submitter.*
/logger_client_log
# Installer logs
pip-log.txt
pip-delete-this-directory.txt
# Unit test / coverage reports
.tox/
.coverage
.cache
.pytest_cache
nosetests.xml
coverage.xml
# Translations
*.mo
# Sphinx documentation
/docs/_build/
*.tar
*.pyc
.idea/
dataset/coco/annotations
dataset/coco/train2017
dataset/coco/val2017
dataset/voc/VOCdevkit
dataset/fruit/fruit-detection/
dataset/voc/test.txt
dataset/voc/trainval.txt
dataset/wider_face/WIDER_test
dataset/wider_face/WIDER_train
dataset/wider_face/WIDER_val
dataset/wider_face/wider_face_split
ppdet/version.py
# NPU meta folder
kernel_meta/
# MAC
*.DS_Store
.pre-commit-config.yaml 0 → 100644
View file @ fccfdfa5
- repo: https://github.com/PaddlePaddle/mirrors-yapf.git
sha: 0d79c0c469bab64f7229c9aca2b1186ef47f0e37
hooks:
- id: yapf
files: \.py$
- repo: https://github.com/pre-commit/pre-commit-hooks
sha: a11d9314b22d8f8c7556443875b731ef05965464
hooks:
- id: check-merge-conflict
- id: check-symlinks
- id: detect-private-key
files: (?!.*paddle)^.*$
- id: end-of-file-fixer
files: \.(md|yml)$
- id: trailing-whitespace
files: \.(md|yml)$
- repo: https://github.com/Lucas-C/pre-commit-hooks
sha: v1.0.1
hooks:
- id: forbid-crlf
files: \.(md|yml)$
- id: remove-crlf
files: \.(md|yml)$
- id: forbid-tabs
files: \.(md|yml)$
- id: remove-tabs
files: \.(md|yml)$
- repo: local
hooks:
- id: clang-format-with-version-check
name: clang-format
description: Format files with ClangFormat.
entry: bash ./.travis/codestyle/clang_format.hook -i
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|proto)$
- repo: local
hooks:
- id: cpplint-cpp-source
name: cpplint
description: Check C++ code style using cpplint.py.
entry: bash ./.travis/codestyle/cpplint_pre_commit.hook
language: system
files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx)$
.style.yapf 0 → 100644
View file @ fccfdfa5
[style]
based_on_style = pep8
column_limit = 80
.travis.yml 0 → 100644
View file @ fccfdfa5
language: cpp
cache: ccache
sudo: required
dist: trusty
services:
- docker
os:
- linux
env:
- JOB=PRE_COMMIT
addons:
apt:
packages:
- git
- python
- python-pip
- python2.7-dev
ssh_known_hosts: 13.229.163.131
before_install:
- sudo pip install -U virtualenv pre-commit pip -i https://pypi.tuna.tsinghua.edu.cn/simple
- docker pull paddlepaddle/paddle:latest
- git pull https://github.com/PaddlePaddle/PaddleDetection develop
script:
- exit_code=0
- .travis/precommit.sh || exit_code=$(( exit_code | $? ))
# - docker run -i --rm -v "$PWD:/py_unittest" paddlepaddle/paddle:latest /bin/bash -c
# 'cd /py_unittest; sh .travis/unittest.sh' || exit_code=$(( exit_code | $? ))
- if [ $exit_code -eq 0 ]; then true; else exit 1; fi;
notifications:
email:
on_success: change
on_failure: always
LICENSE 0 → 100644
View file @ fccfdfa5
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>.
README.md
View file @ fccfdfa5
# Yolov5_paddle # yolov5_paddle
## 论文
无论文
## 模型结构
YOLOv5 是一种目标检测算法，采用单阶段（one-stage）的方法，基于轻量级的卷积神经网络结构，通过引入不同尺度的特征融合和特征金字塔结构来实现高效准确的目标检测。
![Backbone.png](asserts%2FBackbone.png)
## 算法原理
YOLOv5 是一种基于单阶段目标检测算法，通过将图像划分为不同大小的网格，预测每个网格中的目标类别和边界框，利用特征金字塔结构和自适应的模型缩放来实现高效准确的实时目标检测。
![Algorithm_principle.png](asserts%2FAlgorithm_principle.png)
## 环境配置
### Docker (方法一)
```
docker pull image.sourcefind.cn:5000/dcu/admin/base/paddlepaddle:2.4.2-centos7.6-dtk-23.04-py38-latest
docker run -it -v /path/your_code_data/:/path/your_code_data/ --shm-size=32G --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video --name docker_name imageID bash
cd /path/workspace/
pip3 install -r requirements.txt
```
### Dockerfile (方法二)
```
cd ./docker
docker build --no-cache -t yolov5_paddle:last .
docker run -it -v /path/your_code_data/:/path/your_code_data/ --shm-size=32G --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video --name docker_name imageID bash
```
### Anaconda (方法三)
1、关于本项目DCU显卡所需的特殊深度学习库可从光合开发者社区下载安装： https://developer.hpccube.com/tool/
```
DTK软件栈：dtk23.04
python：python3.8
paddle：2.4.2
```
Tips：以上dtk软件栈、python、paddle等DCU相关工具版本需要严格一一对应
2、其他非特殊库直接按照requirements.txt安装
```
pip3 install -r requirements.txt
```
## 数据集
COCO2017（在网络良好的情况下，如果没有下载数据集，程序会默认在线下载数据集）
[训练数据](http://images.cocodataset.org/zips/train2017.zip)
[验证数据](http://images.cocodataset.org/zips/val2017.zip)
[测试数据](http://images.cocodataset.org/zips/test2017.zip)
[标签数据](https://github.com/ultralytics/yolov5/releases/download/v1.0/coco2017labels.zip)
数据集的目录结构如下：
```
├── images
│ ├── train2017
│ ├── val2017
│ ├── test2017
├── labels
│ ├── train2017
│ ├── val2017
├── annotations
│ ├── instances_val2017.json
├── LICENSE
├── README.txt
├── test-dev2017.txt
├── train2017.txt
├── val2017.txt
```
## 训练
### 单机单卡
```
export HIP_VISIBLE_DEVICES=0
export USE_MIOPEN_BATCHNORM=1
python3 train.py --batch 32 --data coco.yaml --cfg 'yolov5m.yaml' --weights '' --project 'run/train' --hyp 'data/hyps/hyp.scratch-high.yaml' --epochs 1000 2>&1 | tee yolov5m.log
```
### 单机多卡
```
#以单机四卡为例子
export HIP_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
export HSA_FORCE_FINE_GRAIN_PCIE=1
export USE_MIOPEN_BATCHNORM=1
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/yolov5/yolov5_s_300e_coco.yml --amp --eval
```
## 推理
#### 单卡推理
```
HIP_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams --infer_img=demo/000000014439_640x640.jpg
```
## result
此处以yolov5s模型进行推理测试
![result_000000014439_640x640.jpg](asserts%2Fresult_000000014439_640x640.jpg)
## 精度
| 模型 | 数据类型 | map0.5:0.95 | map0.5 |
|:-------:|:----:|:-----------:|:------:|
| yolov5s | 单精 | 37.0 | 56.4 |
| yolov5s | 混精 | 37.2 | 56.4 |
## 应用场景
### 算法分类
目标检测
### 热点应用行业
金融,交通,教育
## 源码仓库及问题反馈
https://developer.hpccube.com/codes/modelzoo/yolov5_pytorch
## 参考
[GitHub - ultralytics/yolov5 at v6.0](https://github.com/ultralytics/yolov5/tree/v6.0)
README_cn.md 0 → 100644
View file @ fccfdfa5
简体中文 | [English](README_en.md)
## 简介
**PaddleYOLO**是基于[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)的YOLO系列模型库，**只包含YOLO系列模型的相关代码**，支持`YOLOv3`、`PP-YOLO`、`PP-YOLOv2`、`PP-YOLOE`、**`PP-YOLOE+`**、**`RT-DETR`**、`YOLOX`、`YOLOv5`、`YOLOv6`、`YOLOv7`、`YOLOv8`、`YOLOv5u`、`YOLOv7u`、`YOLOv6Lite`、`RTMDet`等模型，COCO数据集模型库请参照 [ModelZoo](docs/MODEL_ZOO_cn.md) 和 [configs](configs/)。
<div align="center">
<img src="https://user-images.githubusercontent.com/13104100/213197403-c8257486-9ac4-486f-a0d5-4e3fe27ca852.jpg" width="480"/>
<img src="https://user-images.githubusercontent.com/13104100/213197635-eeb55433-bb2d-44f6-b374-73c616cfab24.jpg" width="480"/>
</div>
**注意:**
- **PaddleYOLO** 代码库协议为 **[GPL 3.0](LICENSE)**，[YOLOv5](configs/yolov5)、[YOLOv6](configs/yolov6)、[YOLOv7](configs/yolov7)和[YOLOv8](configs/yolov8)这几类模型代码不合入[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)，其余YOLO模型推荐在[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)中使用，**会最先发布PP-YOLO系列特色检测模型的最新进展**；
- **PaddleYOLO**代码库**推荐使用paddlepaddle-2.4.2以上的版本**，请参考[官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html)下载对应适合版本，**Windows平台请安装paddle develop版本**；
- **PaddleYOLO 的[Roadmap](https://github.com/PaddlePaddle/PaddleYOLO/issues/44)** issue用于收集用户的需求，欢迎提出您的建议和需求；
## 教程
<details open>
<summary>安装</summary>
Clone 代码库和安装 [requirements.txt](./requirements.txt)，环境需要在一个
[**Python>=3.7.0**](https://www.python.org/) 下的环境，且需要安装
[**PaddlePaddle>=2.4.2**](https://www.paddlepaddle.org.cn/install/)。
```bash
git clone https://github.com/PaddlePaddle/PaddleYOLO # clone
cd PaddleYOLO
pip install -r requirements.txt # install
```
</details>
<details open>
<summary>训练/验证/预测/</summary>
将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。
```bash
model_name=ppyoloe # 可修改，如 yolov7
job_name=ppyoloe_plus_crn_s_80e_coco # 可修改，如 yolov7_tiny_300e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 1.训练（单卡/多卡），加 --eval 表示边训边评估，加 --amp 表示混合精度训练
# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp
python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp
# 2.评估，加 --classwise 表示输出每一类mAP
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise
# 3.预测 (单张图/图片文件夹）
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5
# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5
```
</details>
<details>
<summary>部署/测速</summary>
将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。
```bash
model_name=ppyoloe # 可修改，如 yolov7
job_name=ppyoloe_plus_crn_s_80e_coco # 可修改，如 yolov7_tiny_300e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 4.导出模型，以下3种模式选一种
## 普通导出，加trt表示用于trt加速，对NMS和silu激活函数提速明显
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True
## exclude_post_process去除后处理导出，返回和YOLOv5导出ONNX时相同格式的concat后的1个Tensor，是未缩放回原图的坐标+分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True
## exclude_nms去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True
# 5.部署预测，注意不能使用 去除后处理 或 去除NMS 导出后的模型去预测
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU
# 6.部署测速，加 “--run_mode=trt_fp16” 表示在TensorRT FP16模式下测速，注意如需用到 trt_fp16 则必须为加 trt=True 导出的模型
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16
# 7.onnx导出，一般结合 exclude_post_process去除后处理导出的模型
paddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx
# 8.onnx trt测速
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32
```
- 如果想切换模型，只要修改开头两行即可，如:
```
model_name=yolov7
job_name=yolov7_tiny_300e_coco
```
- 导出**onnx**，首先安装[Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX)，`pip install paddle2onnx`；
- **统计FLOPs(G)和Params(M)**，首先安装[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)，`pip install paddleslim`，然后设置[runtime.yml](configs/runtime.yml)里`print_flops: True`和`print_params: True`，并且注意确保是**单尺度**下如640x640，**打印的是MACs，FLOPs=2*MACs**。
</details>
<details open>
<summary> [训练自定义数据集](https://github.com/PaddlePaddle/PaddleYOLO/issues/43) </summary>
- 请参照[文档](docs/MODEL_ZOO_cn.md#自定义数据集)和[issue](https://github.com/PaddlePaddle/PaddleYOLO/issues/43)；
- PaddleDetection团队提供了**基于PP-YOLOE的各种垂类检测模型**的配置文件和权重，用户也可以作为参考去使用自定义数据集。请参考 [PP-YOLOE application](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/ppyoloe/application)、[pphuman](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/pphuman)、[ppvehicle](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/ppvehicle)、[visdrone](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/visdrone) 和 [smalldet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/smalldet)。
- PaddleDetection团队也提供了**VOC数据集的各种YOLO模型**的配置文件和权重，用户也可以作为参考去使用自定义数据集。请参考 [voc](configs/voc)。
- 训练自定义数据集之前请先**确保加载了对应COCO权重作为预训练**，将配置文件中的`pretrain_weights: `设置为对应COCO模型训好的权重，一般会提示head分类层卷积的通道数没对应上，属于正常现象，是由于自定义数据集一般和COCO数据集种类数不一致；
- YOLO检测模型建议**总`batch_size`至少大于`64`**去训练，如果资源不够请**换小模型**或**减小模型的输入尺度**，为了保障较高检测精度，**尽量不要尝试单卡训和总`batch_size`小于`64`训**；
</details>
## 更新日志
* 【2023/05/21】支持[RT-DETR](configs/rtdetr)、[YOLOv8](configs/yolov8)、[YOLOv5u](configs/yolov5/yolov5u)和[YOLOv7u](configs/yolov7/yolov7u)训练全流程，支持[YOLOv6Lite](configs/yolov6/yolov6lite)预测和部署；
* 【2023/03/13】支持[YOLOv5u](configs/yolov5/yolov5u)和[YOLOv7u](configs/yolov7/yolov7u)预测和部署；
* 【2023/01/10】支持[YOLOv8](configs/yolov8)预测和部署；
* 【2022/09/29】支持[RTMDet](configs/rtmdet)预测和部署；
* 【2022/09/26】发布[PaddleYOLO](https://github.com/PaddlePaddle/PaddleYOLO)模型套件，请参照[ModelZoo](docs/MODEL_ZOO_cn.md)；
* 【2022/09/19】支持[YOLOv6](configs/yolov6)新版，包括n/t/s/m/l模型；
* 【2022/08/23】发布`YOLOSeries`代码库: 支持`YOLOv3`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6`,`YOLOv7`等YOLO模型，支持`ConvNeXt`骨干网络高精度版`PP-YOLOE`,`YOLOX`和`YOLOv5`等模型，支持PaddleSlim无损加速量化训练`PP-YOLOE`,`YOLOv5`,`YOLOv6`和`YOLOv7`等模型，详情可阅读[此文章](https://mp.weixin.qq.com/s/Hki01Zs2lQgvLSLWS0btrA)；
## <img src="https://user-images.githubusercontent.com/48054808/157793354-6e7f381a-0aa6-4bb7-845c-9acf2ecc05c3.png" width="20"/> 产品动态
- 🔥 **2023.3.14：PaddleYOLO发布[release/2.6版本](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.6)**
- 💡 模型套件：
- 支持`YOLOv8`,`YOLOv5u`,`YOLOv7u`等YOLO模型预测和部署；
- 支持`Swin-Transformer`、`ViT`、`FocalNet`骨干网络高精度版`PP-YOLOE+`等模型；
- 支持`YOLOv8`在[FastDeploy](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection)中多硬件快速部署；
- 🔥 **2022.9.26：PaddleYOLO发布[release/2.5版本](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.5)**
- 💡 模型套件：
- 发布[PaddleYOLO](https://github.com/PaddlePaddle/PaddleYOLO)模型套件: 支持`YOLOv3`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6`,`YOLOv7`等YOLO模型，支持`ConvNeXt`骨干网络高精度版`PP-YOLOE`,`YOLOX`和`YOLOv5`等模型，支持PaddleSlim无损加速量化训练`PP-YOLOE`,`YOLOv5`,`YOLOv6`和`YOLOv7`等模型；
- 🔥 **2022.8.26：PaddleDetection发布[release/2.5版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5)**
- 🗳 特色模型：
- 发布[PP-YOLOE+](configs/ppyoloe)，最高精度提升2.4% mAP，达到54.9% mAP，模型训练收敛速度提升3.75倍，端到端预测速度最高提升2.3倍；多个下游任务泛化性提升
- 发布[PicoDet-NPU](configs/picodet)模型，支持模型全量化部署；新增[PicoDet](configs/picodet)版面分析模型
- 发布[PP-TinyPose升级版](./configs/keypoint/tiny_pose/)增强版，在健身、舞蹈等场景精度提升9.1% AP，支持侧身、卧躺、跳跃、高抬腿等非常规动作
- 🔮 场景能力：
- 发布行人分析工具[PP-Human v2](./deploy/pipeline)，新增打架、打电话、抽烟、闯入四大行为识别，底层算法性能升级，覆盖行人检测、跟踪、属性三类核心算法能力，提供保姆级全流程开发及模型优化策略，支持在线视频流输入
- 首次发布[PP-Vehicle](./deploy/pipeline)，提供车牌识别、车辆属性分析（颜色、车型）、车流量统计以及违章检测四大功能，兼容图片、在线视频流、视频输入，提供完善的二次开发文档教程
- 💡 前沿算法：
- 全面覆盖的[YOLO家族](https://github.com/PaddlePaddle/PaddleYOLO)经典与最新模型: 包括YOLOv3，百度飞桨自研的实时高精度目标检测检测模型PP-YOLOE，以及前沿检测算法YOLOv4、YOLOv5、YOLOX，YOLOv6及YOLOv7
- 新增基于[ViT](configs/vitdet)骨干网络高精度检测模型，COCO数据集精度达到55.7% mAP；新增[OC-SORT](configs/mot/ocsort)多目标跟踪模型；新增[ConvNeXt](configs/convnext)骨干网络
- 📋 产业范例：新增[智能健身](https://aistudio.baidu.com/aistudio/projectdetail/4385813)、[打架识别](https://aistudio.baidu.com/aistudio/projectdetail/4086987?channelType=0&channel=0)、[来客分析](https://aistudio.baidu.com/aistudio/projectdetail/4230123?channelType=0&channel=0)、车辆结构化范例
- 2022.3.24：PaddleDetection发布[release/2.4版本](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4)
- 发布高精度云边一体SOTA目标检测模型[PP-YOLOE](configs/ppyoloe)，提供s/m/l/x版本，l版本COCO test2017数据集精度51.6%，V100预测速度78.1 FPS，支持混合精度训练，训练较PP-YOLOv2加速33%，全系列多尺度模型，满足不同硬件算力需求，可适配服务器、边缘端GPU及其他服务器端AI加速卡。
- 发布边缘端和CPU端超轻量SOTA目标检测模型[PP-PicoDet增强版](configs/picodet)，精度提升2%左右，CPU预测速度提升63%，新增参数量0.7M的PicoDet-XS模型，提供模型稀疏化和量化功能，便于模型加速，各类硬件无需单独开发后处理模块，降低部署门槛。
- 发布实时行人分析工具[PP-Human](deploy/pipeline)，支持行人跟踪、人流量统计、人体属性识别与摔倒检测四大能力，基于真实场景数据特殊优化，精准识别各类摔倒姿势，适应不同环境背景、光线及摄像角度。
- 新增[YOLOX](configs/yolox)目标检测模型，支持nano/tiny/s/m/l/x版本，x版本COCO val2017数据集精度51.8%。
- [更多版本发布](https://github.com/PaddlePaddle/PaddleDetection/releases)
## <img title="" src="https://user-images.githubusercontent.com/48054808/157795569-9fc77c85-732f-4870-9be0-99a7fe2cff27.png" alt="" width="20"> 简介
**PaddleDetection**为基于飞桨PaddlePaddle的端到端目标检测套件，内置**30+模型算法**及**250+预训练模型**，覆盖**目标检测、实例分割、跟踪、关键点检测**等方向，其中包括**服务器端和移动端高精度、轻量级**产业级SOTA模型、冠军方案和学术前沿算法，并提供配置化的网络模块组件、十余种数据增强策略和损失函数等高阶优化支持和多种部署方案，在打通数据处理、模型开发、训练、压缩、部署全流程的基础上，提供丰富的案例及教程，加速算法产业落地应用。
<div align="center">
<img src="https://user-images.githubusercontent.com/22989727/189026616-75f9c06c-b403-4a61-9372-0fcbed6e0662.gif" width="800"/>
</div>
## <img src="https://user-images.githubusercontent.com/48054808/157799599-e6a66855-bac6-4e75-b9c0-96e13cb9612f.png" width="20"/> 特性
- **模型丰富**: 包含**目标检测**、**实例分割**、**人脸检测**、****关键点检测****、**多目标跟踪**等**250+个预训练模型**，涵盖多种**全球竞赛冠军**方案。
- **使用简洁**：模块化设计，解耦各个网络组件，开发者轻松搭建、试用各种检测模型及优化策略，快速得到高性能、定制化的算法。
- **端到端打通**: 从数据增强、组网、训练、压缩、部署端到端打通，并完备支持**云端**/**边缘端**多架构、多设备部署。
- **高性能**: 基于飞桨的高性能内核，模型训练速度及显存占用优势明显。支持FP16训练, 支持多机训练。
<div align="center">
<img src="https://user-images.githubusercontent.com/22989727/189026189-5d21e93a-5b33-40ce-bc36-c737122c1992.png" width="800"/>
</div>
## <img title="" src="https://user-images.githubusercontent.com/48054808/157800467-2a9946ad-30d1-49a9-b9db-ba33413d9c90.png" alt="" width="20"> 技术交流
- 如果你发现任何PaddleDetection存在的问题或者是建议, 欢迎通过[GitHub Issues](https://github.com/PaddlePaddle/PaddleDetection/issues)给我们提issues。
- **欢迎加入PaddleDetection 微信用户群（扫码填写问卷即可入群）**
- **入群福利 💎：获取PaddleDetection团队整理的重磅学习大礼包🎁**
- 📊 福利一：获取飞桨联合业界企业整理的开源数据集
- 👨‍🏫 福利二：获取PaddleDetection历次发版直播视频与最新直播咨询
- 🗳 福利三：获取垂类场景预训练模型集合，包括工业、安防、交通等5+行业场景
- 🗂 福利四：获取10+全流程产业实操范例，覆盖火灾烟雾检测、人流量计数等产业高频场景
<div align="center">
<img src="https://user-images.githubusercontent.com/34162360/177678712-4655747d-4290-4ad9-b7a1-4564a5418ac6.jpg" width = "200" />
</div>
## <img src="https://user-images.githubusercontent.com/48054808/157827140-03ffaff7-7d14-48b4-9440-c38986ea378c.png" width="20"/> 套件结构概览
<table align="center">
<tbody>
<tr align="center" valign="bottom">
<td>
<b>Architectures</b>
</td>
<td>
<b>Backbones</b>
</td>
<td>
<b>Components</b>
</td>
<td>
<b>Data Augmentation</b>
</td>
</tr>
<tr valign="top">
<td>
<ul>
<details open><summary><b>Object Detection</b></summary>
<ul>
<li>YOLOv3</li>
<li>YOLOv5</li>
<li>YOLOv6</li>
<li>YOLOv7</li>
<li>YOLOv8</li>
<li>PP-YOLOv1/v2</li>
<li>PP-YOLO-Tiny</li>
<li>PP-YOLOE</li>
<li>PP-YOLOE+</li>
<li>YOLOX</li>
<li>RTMDet</li>
</ul></details>
</ul>
</td>
<td>
<details open><summary><b>Details</b></summary>
<ul>
<li>ResNet(&vd)</li>
<li>CSPResNet</li>
<li>DarkNet</li>
<li>CSPDarkNet</li>
<li>ConvNeXt</li>
<li>EfficientRep</li>
<li>CSPBepBackbone</li>
<li>ELANNet</li>
<li>CSPNeXt</li>
</ul></details>
</td>
<td>
<details open><summary><b>Common</b></summary>
<ul>
<li>Sync-BN</li>
<li>Group Norm</li>
<li>DCNv2</li>
<li>EMA</li>
</ul> </details>
</ul>
<details open><summary><b>FPN</b></summary>
<ul>
<li>YOLOv3FPN</li>
<li>PPYOLOFPN</li>
<li>PPYOLOTinyFPN</li>
<li>PPYOLOPAN</li>
<li>YOLOCSPPAN</li>
<li>Custom-PAN</li>
<li>RepPAN</li>
<li>CSPRepPAN</li>
<li>ELANFPN</li>
<li>ELANFPNP6</li>
<li>CSPNeXtPAFPN</li>
</ul> </details>
</ul>
<details open><summary><b>Loss</b></summary>
<ul>
<li>Smooth-L1</li>
<li>GIoU/DIoU/CIoU</li>
<li>IoUAware</li>
<li>Focal Loss</li>
<li>VariFocal Loss</li>
</ul> </details>
</ul>
<details open><summary><b>Post-processing</b></summary>
<ul>
<li>SoftNMS</li>
<li>MatrixNMS</li>
</ul> </details>
</ul>
<details open><summary><b>Speed</b></summary>
<ul>
<li>FP16 training</li>
<li>Multi-machine training </li>
</ul> </details>
</ul>
</td>
<td>
<details open><summary><b>Details</b></summary>
<ul>
<li>Resize</li>
<li>Lighting</li>
<li>Flipping</li>
<li>Expand</li>
<li>Crop</li>
<li>Color Distort</li>
<li>Random Erasing</li>
<li>Mixup </li>
<li>AugmentHSV</li>
<li>Mosaic</li>
<li>Cutmix </li>
<li>Grid Mask</li>
<li>Auto Augment</li>
<li>Random Perspective</li>
</ul> </details>
</td>
</tr>
</td>
</tr>
</tbody>
</table>
## <img src="https://user-images.githubusercontent.com/48054808/157801371-9a9a8c65-1690-4123-985a-e0559a7f9494.png" width="20"/> 模型性能概览
<details>
<summary><b> 云端模型性能对比</b></summary>
各模型结构和骨干网络的代表模型在COCO数据集上精度mAP和单卡Tesla V100上预测速度(FPS)对比图。
<div align="center">
<img src="docs/images/fps_map.png" />
</div>
**说明：**
- `PP-YOLOE`是对`PP-YOLO v2`模型的进一步优化，在COCO数据集精度51.6%，Tesla V100预测速度78.1FPS
- `PP-YOLOE+`是对`PPOLOE`模型的进一步优化，在COCO数据集精度53.3%，Tesla V100预测速度78.1FPS
- 图中模型均可在[模型库](#模型库)中获取
</details>
<details>
<summary><b> 移动端模型性能对比</b></summary>
各移动端模型在COCO数据集上精度mAP和高通骁龙865处理器上预测速度(FPS)对比图。
<div align="center">
<img src="docs/images/mobile_fps_map.png" width=600/>
</div>
**说明：**
- 测试数据均使用高通骁龙865(4\*A77 + 4\*A55)处理器batch size为1, 开启4线程测试，测试使用NCNN预测库，测试脚本见[MobileDetBenchmark](https://github.com/JiweiMaster/MobileDetBenchmark)
- [PP-PicoDet](configs/picodet)及[PP-YOLO-Tiny](configs/ppyolo)为PaddleDetection自研模型，其余模型PaddleDetection暂未提供
</details>
## <img src="https://user-images.githubusercontent.com/48054808/157829890-a535b8a6-631c-4c87-b861-64d4b32b2d6a.png" width="20"/> 模型库
<details>
<summary><b> 1. 通用检测</b></summary>
#### [PP-YOLOE+](./configs/ppyoloe)系列 推荐场景：Nvidia V100, T4等云端GPU和Jetson系列等边缘端设备
| 模型名称 | COCO精度（mAP） | V100 TensorRT FP16速度(FPS) | 配置文件 | 模型下载 |
|:---------- |:-----------:|:-------------------------:|:-----------------------------------------------------:|:------------------------------------------------------------------------------------:|
| PP-YOLOE+_s | 43.9 | 333.3 | [链接](configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_s_80e_coco.pdparams) |
| PP-YOLOE+_m | 50.0 | 208.3 | [链接](configs/ppyoloe/ppyoloe_plus_crn_m_80e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_m_80e_coco.pdparams) |
| PP-YOLOE+_l | 53.3 | 149.2 | [链接](configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_m_80e_coco.pdparams) |
| PP-YOLOE+_x | 54.9 | 95.2 | [链接](configs/ppyoloe/ppyoloe_plus_crn_x_80e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_x_80e_coco.pdparams) |
#### 前沿检测算法
| 模型名称 | COCO精度（mAP） | V100 TensorRT FP16速度(FPS) | 配置文件 | 模型下载 |
|:------------------------------------------------------------------ |:-----------:|:-------------------------:|:------------------------------------------------------------------------------------------------------------:|:--------------------------------------------------------------------------:|
| [YOLOX-l](configs/yolox) | 50.1 | 107.5 | [链接](configs/yolox/yolox_l_300e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/yolox_l_300e_coco.pdparams) |
| [YOLOv5-l](configs/yolov5) | 48.6 | 136.0 | [链接](configs/yolov5/yolov5_l_300e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/yolov5_l_300e_coco.pdparams) |
| [YOLOv7-l](configs/yolov7) | 51.0 | 135.0 | [链接](configs/yolov7/yolov7_l_300e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/yolov7_l_300e_coco.pdparams) |
</details>
## <img src="https://user-images.githubusercontent.com/48054808/157828296-d5eb0ccb-23ea-40f5-9957-29853d7d13a9.png" width="20"/> 文档教程
### 入门教程
- [安装说明](docs/tutorials/INSTALL_cn.md)
- [快速体验](docs/tutorials/QUICK_STARTED_cn.md)
- [数据准备](docs/tutorials/data/README.md)
- [PaddleDetection全流程使用](docs/tutorials/GETTING_STARTED_cn.md)
- [FAQ/常见问题汇总](docs/tutorials/FAQ)
### 进阶教程
- 参数配置
- [PP-YOLO参数说明](docs/tutorials/config_annotation/ppyolo_r50vd_dcn_1x_coco_annotation.md)
- 模型压缩(基于[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim))
- [剪裁/量化/蒸馏教程](configs/slim)
- [推理部署](deploy/README.md)
- [模型导出教程](deploy/EXPORT_MODEL.md)
- [Paddle Inference部署](deploy/README.md)
- [Python端推理部署](deploy/python)
- [C++端推理部署](deploy/cpp)
- [Paddle-Lite部署](deploy/lite)
- [Paddle Serving部署](deploy/serving)
- [ONNX模型导出](deploy/EXPORT_ONNX_MODEL.md)
- [推理benchmark](deploy/BENCHMARK_INFER.md)
- 进阶开发
- [数据处理模块](docs/advanced_tutorials/READER.md)
- [新增检测模型](docs/advanced_tutorials/MODEL_TECHNICAL.md)
- 二次开发教程
- [目标检测](docs/advanced_tutorials/customization/detection.md)
## <img src="https://user-images.githubusercontent.com/48054808/157835981-ef6057b4-6347-4768-8fcc-cd07fcc3d8b0.png" width="20"/> 版本更新
版本更新内容请参考[版本更新文档](docs/CHANGELOG.md)
## <img title="" src="https://user-images.githubusercontent.com/48054808/157835345-f5d24128-abaf-4813-b793-d2e5bdc70e5a.png" alt="" width="20"> 许可证书
本项目的发布受[GPL-3.0 license](LICENSE)许可认证。
## <img src="https://user-images.githubusercontent.com/48054808/157835276-9aab9d1c-1c46-446b-bdd4-5ab75c5cfa48.png" width="20"/> 引用
```
@misc{ppdet2019,
title={PaddleDetection, Object detection and instance segmentation toolkit based on PaddlePaddle.},
author={PaddlePaddle Authors},
howpublished = {\url{https://github.com/PaddlePaddle/PaddleDetection}},
year={2019}
}
```
README_en.md 0 → 100644
View file @ fccfdfa5
[简体中文](README_cn.md) | English
## Introduction
**PaddleYOLO** is a YOLO series toolbox based on [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection), **only relevant codes of YOLO series models are included**. It supports `YOLOv3`,`PP-YOLO`,`PP-YOLOv2`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6`,`YOLOv7`,`YOLOv8`,`YOLOv5u`,`YOLOv7u`,`RTMDet` and so on, see COCO dataset ModelZoo in [ModelZoo](docs/MODEL_ZOO_en.md) and [configs](configs/).
<div align="center">
<img src="https://user-images.githubusercontent.com/13104100/213197403-c8257486-9ac4-486f-a0d5-4e3fe27ca852.jpg" width="480"/>
<img src="https://user-images.githubusercontent.com/13104100/213197635-eeb55433-bb2d-44f6-b374-73c616cfab24.jpg" width="480"/>
</div>
**Notes：**
- The Licence of **PaddleYOLO** is **[GPL 3.0](LICENSE)**, the codes of [YOLOv5](configs/yolov5),[YOLOv6](configs/yolov6),[YOLOv7](configs/yolov7) and [YOLOv8](configs/yolov8) will not be merged into [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection). Except for these YOLO models, other YOLO models are recommended to use in [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection), **which will be the first to release the latest progress of PP-YOLO series detection model**;
- To use **PaddleYOLO**, **PaddlePaddle-2.3.2 or above is recommended**，please refer to the [official website](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html) to download the appropriate version. **For Windows platforms, please install the paddle develop version**;
- **PaddleYOLO's [Roadmap](https://github.com/PaddlePaddle/PaddleYOLO/issues/44)** issue collects feature requests from user, welcome to put forward any opinions and suggestions.
## Tutorials
<details open>
<summary>Install</summary>
Clone repo and install [requirements.txt](./requirements.txt) in a
[**Python>=3.7.0**](https://www.python.org/) environment, including
[**PaddlePaddle>=2.3.2**](https://www.paddlepaddle.org.cn/install/).
```bash
git clone https://github.com/PaddlePaddle/PaddleYOLO # clone
cd PaddleYOLO
pip install -r requirements.txt # install
```
</details>
<details open>
<summary>Training/Evaluation/Inference</summary>
Write the following commands in a script file, such as ```run.sh```, and run as：```sh run.sh```. You can also run the command line sentence by sentence.
```bash
model_name=ppyoloe # yolov7
job_name=ppyoloe_plus_crn_s_80e_coco # yolov7_tiny_300e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 1.training（single GPU / multi GPU）
# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp
python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp
# 2.eval
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise
# 3.infer
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5
# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5
```
</details>
<details>
<summary>Deployment/Speed</summary>
Write the following commands in a script file, such as ```run.sh```, and run as：```sh run.sh```. You can also run the command line sentence by sentence.
```bash
model_name=ppyoloe # yolov7
job_name=ppyoloe_plus_crn_s_80e_coco # yolov7_tiny_300e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 4.export
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True
# 5.deploy infer
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU
# 6.deploy speed, add '--run_mode=trt_fp16' to test in TensorRT FP16 mode
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16
# 7.export onnx
paddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx
# 8.onnx speed
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32
```
**Note：**
- If you want to switch models, just modify the first two lines, such as:
```
model_name=yolov7
job_name=yolov7_tiny_300e_coco
```
- For **exporting onnx**, you should install [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX) by `pip install paddle2onnx` at first.
- For **FLOPs(G) and Params(M)**, you should install [PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim) by `pip install paddleslim` at first, then set `print_flops: True` and `print_params: True` in [runtime.yml](configs/runtime.yml). Make sure **single scale** like 640x640, **MACs are printed，FLOPs=2*MACs**.
</details>
<details open>
<summary> [Training Custom dataset](https://github.com/PaddlePaddle/PaddleYOLO/issues/43) </summary>
- Please refer to [doc](docs/MODEL_ZOO_en.md#CustomDataset) and [issue](https://github.com/PaddlePaddle/PaddleYOLO/issues/43).
- PaddleDetection team provides various **feature detection models based on PP-YOLOE** , which can also be used as a reference to modify on your custom dataset. Please refer to [PP-YOLOE application](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/ppyoloe/application), [pphuman](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/pphuman), [ppvehicle](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/ppvehicle), [visdrone](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/visdrone) and [smalldet](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.6/configs/smalldet).
- PaddleDetection also provides **various YOLO models for VOC dataset** , which can also be used as a reference to modify on your custom dataset. Please refer to [voc](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.6/configs/voc).
- Please **ensure the corresponding COCO trained weights are loaded as pre-train weights** at first. Set the `pretrain_weights: ` with corresponding COCO trained weights in the config file, and it will generally prompt that the number of channels convolved by the head classification layer does not correspond, which is a normal phenomenon, because the number of types of user-defined data sets is generally inconsistent with that of COCO data sets.
- We recommend to use YOLO detection model **with a total `batch_size` at least greater than `64` to train**. If the resources are insufficient, please **use the smaller model** or **reduce the input size of the model**. To ensure high detection accuracy, **you'd better not try to using single GPU or total `batch_size` less than `64` for training**;
</details>
## Updates
* 【2023/03/13】Support [YOLOv5u](configs/yolov5/yolov5u) and [YOLOv7u](configs/yolov7/yolov7u) inference and deploy;
* 【2023/01/10】Support [YOLOv8](configs/yolov8) inference and deploy;
* 【2022/09/29】Support [RTMDet](configs/rtmdet) inference and deploy;
* 【2022/09/26】Release [PaddleYOLO](https://github.com/PaddlePaddle/PaddleYOLO), see [ModelZoo](docs/MODEL_ZOO_en.md);
* 【2022/09/19】Support the new version of [YOLOv6](configs/yolov6), including n/t/s/m/l model;
* 【2022/08/23】Release `YOLOSeries` codebase: support `YOLOv3`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6` and `YOLOv7`; support using `ConvNeXt` backbone to get high-precision version of `PP-YOLOE`,`YOLOX` and `YOLOv5`; support PaddleSlim accelerated quantitative training `PP-YOLOE`,`YOLOv5`,`YOLOv6` and `YOLOv7`. For details, please read this [article](https://mp.weixin.qq.com/s/Hki01Zs2lQgvLSLWS0btrA)；
## <img src="https://user-images.githubusercontent.com/48054808/157793354-6e7f381a-0aa6-4bb7-845c-9acf2ecc05c3.png" width="20"/> Product Update
- 🔥 **2023.3.14：Release PaddleYOLO [release/2.6](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.6)**
- 💡 Model kit：
- Support `YOLOv8`,`YOLOv5u`,`YOLOv7u` inference and deploy.
- Support `Swin-Transformer`、`ViT`、`FocalNet` backbone to get high-precision version of `PP-YOLOE+`.
- Support `YOLOv8` in [FastDeploy](https://github.com/PaddlePaddle/FastDeploy/tree/develop/examples/vision/detection/paddledetection).
- 🔥 **2022.9.26：Release PaddleYOLO [release/2.5](https://github.com/PaddlePaddle/PaddleYOLO/tree/release/2.5)**
- 💡 Model kit：
- Release [PaddleYOLO](https://github.com/PaddlePaddle/PaddleYOLO): support `YOLOv3`,`PP-YOLOE`,`PP-YOLOE+`,`YOLOX`,`YOLOv5`,`YOLOv6` and `YOLOv7`; support using `ConvNeXt` backbone to get high-precision version of `PP-YOLOE`,`YOLOX` and `YOLOv5`; support PaddleSlim accelerated quantitative training `PP-YOLOE`,`YOLOv5`,`YOLOv6` and `YOLOv7`.
- 🔥 **2022.8.26：PaddleDetection [release/2.5 version](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5)**
- 🗳 Model features：
- Release [PP-YOLOE+](configs/ppyoloe): Increased accuracy by a maximum of 2.4% mAP to 54.9% mAP, 3.75 times faster model training convergence rate, and up to 2.3 times faster end-to-end inference speed; improved generalization for multiple downstream tasks
- Release [PicoDet-NPU](configs/picodet) model which supports full quantization deployment of models; add [PicoDet](configs/picodet) layout analysis model
- Release [PP-TinyPose Plus](./configs/keypoint/tiny_pose/). With 9.1% AP accuracy improvement in physical exercise, dance, and other scenarios, our PP-TinyPose Plus supports unconventional movements such as turning to one side, lying down, jumping, and high lifts
- 🔮 Functions in different scenarios
- Release the pedestrian analysis tool [PP-Human v2](./deploy/pipeline). It introduces four new behavior recognition: fighting, telephoning, smoking, and trespassing. The underlying algorithm performance is optimized, covering three core algorithm capabilities: detection, tracking, and attributes of pedestrians. Our model provides end-to-end development and model optimization strategies for beginners and supports online video streaming input.
- First release [PP-Vehicle](./deploy/pipeline), which has four major functions: license plate recognition, vehicle attribute analysis (color, model), traffic flow statistics, and violation detection. It is compatible with input formats, including pictures, online video streaming, and video. And we also offer our users a comprehensive set of tutorials for customization.
- 💡 Cutting-edge algorithms：
- Covers [YOLO family](https://github.com/PaddlePaddle/PaddleYOLO) classic and latest models: YOLOv3, PP-YOLOE (a real-time high-precision object detection model developed by Baidu PaddlePaddle), and cutting-edge detection algorithms such as YOLOv4, YOLOv5, YOLOX, YOLOv6, and YOLOv7
- Newly add high precision detection model based on [ViT](configs/vitdet) backbone network, with a 55.7% mAP accuracy on COCO dataset; newly add multi-object tracking model [OC-SORT](configs/mot/ocsort); newly add [ConvNeXt](configs/convnext) backbone network.
- 📋 Industrial applications: Newly add [Smart Fitness](https://aistudio.baidu.com/aistudio/projectdetail/4385813), [Fighting recognition](https://aistudio.baidu.com/aistudio/projectdetail/4086987?channelType=0&channel=0),[ and Visitor Analysis](https://aistudio.baidu.com/aistudio/projectdetail/4230123?channelType=0&channel=0).
- 2022.3.24：PaddleDetection released[release/2.4 version](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.4)
- Release high-performanace SOTA object detection model [PP-YOLOE](configs/ppyoloe). It integrates cloud and edge devices and provides S/M/L/X versions. In particular, Verson L has the accuracy as 51.4% on COCO test 2017 dataset, inference speed as 78.1 FPS on a single Test V100. It supports mixed precision training, 33% faster than PP-YOLOv2. Its full range of multi-sized models can meet different hardware arithmetic requirements, and adaptable to server, edge-device GPU and other AI accelerator cards on servers.
- Release ultra-lightweight SOTA object detection model [PP-PicoDet Plus](configs/picodet) with 2% improvement in accuracy and 63% improvement in CPU inference speed. Add PicoDet-XS model with a 0.7M parameter, providing model sparsification and quantization functions for model acceleration. No specific post processing module is required for all the hardware, simplifying the deployment.
- Release the real-time pedestrian analysis tool [PP-Human](deploy/pphuman). It has four major functions: pedestrian tracking, visitor flow statistics, human attribute recognition and falling detection. For falling detection, it is optimized based on real-life data with accurate recognition of various types of falling posture. It can adapt to different environmental background, light and camera angle.
- Add [YOLOX](configs/yolox) object detection model with nano/tiny/S/M/L/X. X version has the accuracy as 51.8% on COCO Val2017 dataset.
- [More releases](https://github.com/PaddlePaddle/PaddleDetection/releases)
## <img title="" src="https://user-images.githubusercontent.com/48054808/157795569-9fc77c85-732f-4870-9be0-99a7fe2cff27.png" alt="" width="20"> Brief Introduction
**PaddleDetection** is an end-to-end object detection development kit based on PaddlePaddle. Providing **over 30 model algorithm** and **over 250 pre-trained models**, it covers object detection, instance segmentation, keypoint detection, multi-object tracking. In particular, PaddleDetection offers **high- performance & light-weight** industrial SOTA models on **servers and mobile** devices, champion solution and cutting-edge algorithm. PaddleDetection provides various data augmentation methods, configurable network components, loss functions and other advanced optimization & deployment schemes. In addition to running through the whole process of data processing, model development, training, compression and deployment, PaddlePaddle also provides rich cases and tutorials to accelerate the industrial application of algorithm.
<div align="center">
<img src="https://user-images.githubusercontent.com/22989727/189122825-ee1c1db2-b5f9-42c0-88b4-7975e1ec239d.gif" width="800"/>
</div>
## <img src="https://user-images.githubusercontent.com/48054808/157799599-e6a66855-bac6-4e75-b9c0-96e13cb9612f.png" width="20"/> Features
- **Rich model library**: PaddleDetection provides over 250 pre-trained models including **object detection, instance segmentation, face recognition, multi-object tracking**. It covers a variety of **global competition champion** schemes.
- **Simple to use**: Modular design, decoupling each network component, easy for developers to build and try various detection models and optimization strategies, quick access to high-performance, customized algorithm.
- **Getting Through End to End**: PaddlePaddle gets through end to end from data augmentation, constructing models, training, compression, depolyment. It also supports multi-architecture, multi-device deployment for **cloud and edge** device.
- **High Performance**: Due to the high performance core, PaddlePaddle has clear advantages in training speed and memory occupation. It also supports FP16 training and multi-machine training.
<div align="center">
<img src="https://user-images.githubusercontent.com/22989727/189066615-89d1dde2-54bc-4946-887e-fce50069206e.png" width="800"/>
</div>
## <img title="" src="https://user-images.githubusercontent.com/48054808/157800467-2a9946ad-30d1-49a9-b9db-ba33413d9c90.png" alt="" width="20"> Exchanges
- If you have any question or suggestion, please give us your valuable input via [GitHub Issues](https://github.com/PaddlePaddle/PaddleDetection/issues)
Welcome to join PaddleDetection user groups on WeChat (scan the QR code, add and reply "D" to the assistant)
<div align="center">
<img src="https://user-images.githubusercontent.com/34162360/177678712-4655747d-4290-4ad9-b7a1-4564a5418ac6.jpg" width = "200" />
</div>
## <img src="https://user-images.githubusercontent.com/48054808/157827140-03ffaff7-7d14-48b4-9440-c38986ea378c.png" width="20"/> Kit Structure
<table align="center">
<tbody>
<tr align="center" valign="bottom">
<td>
<b>Architectures</b>
</td>
<td>
<b>Backbones</b>
</td>
<td>
<b>Components</b>
</td>
<td>
<b>Data Augmentation</b>
</td>
</tr>
<tr valign="top">
<td>
<ul>
<details open><summary><b>Object Detection</b></summary>
<ul>
<li>YOLOv3</li>
<li>YOLOv5</li>
<li>YOLOv6</li>
<li>YOLOv7</li>
<li>YOLOv8</li>
<li>PP-YOLOv1/v2</li>
<li>PP-YOLO-Tiny</li>
<li>PP-YOLOE</li>
<li>PP-YOLOE+</li>
<li>YOLOX</li>
<li>RTMDet</li>
</ul></details>
</ul>
</td>
<td>
<details open><summary><b>Details</b></summary>
<ul>
<li>ResNet(&vd)</li>
<li>CSPResNet</li>
<li>DarkNet</li>
<li>CSPDarkNet</li>
<li>ConvNeXt</li>
<li>EfficientRep</li>
<li>CSPBepBackbone</li>
<li>ELANNet</li>
<li>CSPNeXt</li>
</ul></details>
</td>
<td>
<details open><summary><b>Common</b></summary>
<ul>
<li>Sync-BN</li>
<li>Group Norm</li>
<li>DCNv2</li>
<li>EMA</li>
</ul> </details>
</ul>
<details open><summary><b>FPN</b></summary>
<ul>
<li>YOLOv3FPN</li>
<li>PPYOLOFPN</li>
<li>PPYOLOTinyFPN</li>
<li>PPYOLOPAN</li>
<li>YOLOCSPPAN</li>
<li>Custom-PAN</li>
<li>RepPAN</li>
<li>CSPRepPAN</li>
<li>ELANFPN</li>
<li>ELANFPNP6</li>
<li>CSPNeXtPAFPN</li>
</ul> </details>
</ul>
<details open><summary><b>Loss</b></summary>
<ul>
<li>Smooth-L1</li>
<li>GIoU/DIoU/CIoU</li>
<li>IoUAware</li>
<li>Focal Loss</li>
<li>VariFocal Loss</li>
</ul> </details>
</ul>
<details open><summary><b>Post-processing</b></summary>
<ul>
<li>SoftNMS</li>
<li>MatrixNMS</li>
</ul> </details>
</ul>
<details open><summary><b>Speed</b></summary>
<ul>
<li>FP16 training</li>
<li>Multi-machine training </li>
</ul> </details>
</ul>
</td>
<td>
<details open><summary><b>Details</b></summary>
<ul>
<li>Resize</li>
<li>Lighting</li>
<li>Flipping</li>
<li>Expand</li>
<li>Crop</li>
<li>Color Distort</li>
<li>Random Erasing</li>
<li>Mixup </li>
<li>AugmentHSV</li>
<li>Mosaic</li>
<li>Cutmix </li>
<li>Grid Mask</li>
<li>Auto Augment</li>
<li>Random Perspective</li>
</ul> </details>
</td>
</tr>
</td>
</tr>
</tbody>
</table>
## <img src="https://user-images.githubusercontent.com/48054808/157801371-9a9a8c65-1690-4123-985a-e0559a7f9494.png" width="20"/> Model Performance
<details>
<summary><b> Performance comparison of Cloud models</b></summary>
The comparison between COCO mAP and FPS on Tesla V100 of representative models of each architectures and backbones.
<div align="center">
<img src="docs/images/fps_map.png" />
</div>
**Clarification：**
- `PP-YOLOE` are optimized `PP-YOLO v2`. It reached accuracy as 51.4% on COCO dataset, inference speed as 78.1 FPS on Tesla V100
- `PP-YOLOE+` are optimized `PP-YOLOE`. It reached accuracy as 53.3% on COCO dataset, inference speed as 78.1 FPS on Tesla V100
- The models in the figure are available in the[ model library](#模型库)
</details>
<details>
<summary><b> Performance omparison on mobiles</b></summary>
The comparison between COCO mAP and FPS on Qualcomm Snapdragon 865 processor of models on mobile devices.
<div align="center">
<img src="docs/images/mobile_fps_map.png" width=600/>
</div>
**Clarification：**
- Tests were conducted on Qualcomm Snapdragon 865 (4 \*A77 + 4 \*A55) batch_size=1, 4 thread, and NCNN inference library, test script see [MobileDetBenchmark](https://github.com/JiweiMaster/MobileDetBenchmark)
- [PP-PicoDet](configs/picodet) and [PP-YOLO-Tiny](configs/ppyolo) are self-developed models of PaddleDetection, and other models are not tested yet.
</details>
## <img src="https://user-images.githubusercontent.com/48054808/157829890-a535b8a6-631c-4c87-b861-64d4b32b2d6a.png" width="20"/> Model libraries
<details>
<summary><b> 1. General detection</b></summary>
#### PP-YOLOE series Recommended scenarios: Cloud GPU such as Nvidia V100, T4 and edge devices such as Jetson series
| Model | COCO Accuracy（mAP） | V100 TensorRT FP16 Speed(FPS) | Configuration | Download |
|:---------- |:------------------:|:-----------------------------:|:-------------------------------------------------------:|:----------------------------------------------------------------------------------------:|
| PP-YOLOE+_s | 43.9 | 333.3 | [link](configs/ppyoloe/ppyoloe_plus_crn_s_80e_coco.yml) | [download](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_s_80e_coco.pdparams) |
| PP-YOLOE+_m | 50.0 | 208.3 | [link](configs/ppyoloe/ppyoloe_plus_crn_m_80e_coco.yml) | [download](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_m_80e_coco.pdparams) |
| PP-YOLOE+_l | 53.3 | 149.2 | [link](configs/ppyoloe/ppyoloe_plus_crn_l_80e_coco.yml) | [download](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_m_80e_coco.pdparams) |
| PP-YOLOE+_x | 54.9 | 95.2 | [link](configs/ppyoloe/ppyoloe_plus_crn_x_80e_coco.yml) | [download](https://paddledet.bj.bcebos.com/models/ppyoloe_plus_crn_x_80e_coco.pdparams) |
#### Frontier detection algorithm
| Model | COCO Accuracy（mAP） | V100 TensorRT FP16 speed(FPS) | Configuration | Download |
|:-------- |:------------------:|:-----------------------------:|:--------------------------------------------------------------------------------------------------------------:|:------------------------------------------------------------------------------:|
| [YOLOX-l](configs/yolox) | 50.1 | 107.5 | [Link](configs/yolox/yolox_l_300e_coco.yml) | [Download](https://paddledet.bj.bcebos.com/models/yolox_l_300e_coco.pdparams) |
| [YOLOv5-l](configs/yolov5) | 48.6 | 136.0 | [Link](configs/yolov5/yolov5_l_300e_coco.yml) | [Download](https://paddledet.bj.bcebos.com/models/yolov5_l_300e_coco.pdparams) |
| [YOLOv7-l](configs/yolov7) | 51.0 | 135.0 | [链接](configs/yolov7/yolov7_l_300e_coco.yml) | [下载地址](https://paddledet.bj.bcebos.com/models/yolov7_l_300e_coco.pdparams) |
</details>
## <img src="https://user-images.githubusercontent.com/48054808/157828296-d5eb0ccb-23ea-40f5-9957-29853d7d13a9.png" width="20"/>Document tutorials
### Introductory tutorials
- [Installation](docs/tutorials/INSTALL_cn.md)
- [Quick start](docs/tutorials/QUICK_STARTED_cn.md)
- [Data preparation](docs/tutorials/data/README.md)
- [Geting Started on PaddleDetection](docs/tutorials/GETTING_STARTED_cn.md)
- [FAQ]((docs/tutorials/FAQ)
### Advanced tutorials
- Configuration
- [PP-YOLO Configuration](docs/tutorials/config_annotation/ppyolo_r50vd_dcn_1x_coco_annotation.md)
- Compression based on [PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)
- [Pruning/Quantization/Distillation Tutorial](configs/slim)
- [Inference deployment](deploy/README.md)
- [Export model for inference](deploy/EXPORT_MODEL.md)
- [Paddle Inference deployment](deploy/README.md)
- [Inference deployment with Python](deploy/python)
- [Inference deployment with C++](deploy/cpp)
- [Paddle-Lite deployment](deploy/lite)
- [Paddle Serving deployment](deploy/serving)
- [ONNX model export](deploy/EXPORT_ONNX_MODEL.md)
- [Inference benchmark](deploy/BENCHMARK_INFER.md)
- Advanced development
- [Data processing module](docs/advanced_tutorials/READER.md)
- [New object detection models](docs/advanced_tutorials/MODEL_TECHNICAL.md)
- Custumization
- [Object detection](docs/advanced_tutorials/customization/detection.md)
## <img src="https://user-images.githubusercontent.com/48054808/157835981-ef6057b4-6347-4768-8fcc-cd07fcc3d8b0.png" width="20"/> Version updates
Please refer to the[ Release note ](https://github.com/PaddlePaddle/Paddle/wiki/PaddlePaddle-2.3.0-Release-Note-EN)for more details about the updates
## <img title="" src="https://user-images.githubusercontent.com/48054808/157835345-f5d24128-abaf-4813-b793-d2e5bdc70e5a.png" alt="" width="20"> License
PaddleYOLO is provided under the [GPL-3.0 license](LICENSE)
## <img src="https://user-images.githubusercontent.com/48054808/157835276-9aab9d1c-1c46-446b-bdd4-5ab75c5cfa48.png" width="20"/> Quote
```
@misc{ppdet2019,
title={PaddleDetection, Object detection and instance segmentation toolkit based on PaddlePaddle.},
author={PaddlePaddle Authors},
howpublished = {\url{https://github.com/PaddlePaddle/PaddleDetection}},
year={2019}
}
```
asserts/model.properties 0 → 100644
View file @ fccfdfa5
# 模型唯一标识
modelCode = 469
# 模型名称
modelName=yolov5_paddle
# 模型描述
modelDescription=yolov5是一种基于深度学习的目标检测算法,可以广泛应用于各种计算机视觉和人工智能领域的应用中
# 应用场景
appScenario=推理,训练,金融,交通,教育
# 框架类型
frameType=paddle
\ No newline at end of file
configs/datasets/coco_detection.yml 0 → 100644
View file @ fccfdfa5
metric: COCO
num_classes: 80
TrainDataset:
name: COCODataSet
image_dir: train2017
anno_path: annotations/instances_train2017.json
dataset_dir: dataset/coco
data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
EvalDataset:
name: COCODataSet
image_dir: val2017
anno_path: annotations/instances_val2017.json
dataset_dir: dataset/coco
TestDataset:
name: ImageFolder
anno_path: annotations/instances_val2017.json # also support txt (like VOC's label_list.txt)
dataset_dir: dataset/coco # if set, anno_path will be 'dataset_dir/anno_path'
configs/datasets/objects365_detection.yml 0 → 100644
View file @ fccfdfa5
metric: COCO
num_classes: 365
TrainDataset:
!COCODataSet
image_dir: train
anno_path: annotations/zhiyuan_objv2_train.json
dataset_dir: dataset/objects365
data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
EvalDataset:
!COCODataSet
image_dir: val
anno_path: annotations/zhiyuan_objv2_val.json
dataset_dir: dataset/objects365
allow_empty: true
TestDataset:
!ImageFolder
anno_path: annotations/zhiyuan_objv2_val.json
dataset_dir: dataset/objects365/
configs/datasets/roadsign_voc.yml 0 → 100644
View file @ fccfdfa5
metric: VOC
map_type: integral
num_classes: 4
TrainDataset:
name: VOCDataSet
dataset_dir: dataset/roadsign_voc
anno_path: train.txt
label_list: label_list.txt
data_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']
EvalDataset:
name: VOCDataSet
dataset_dir: dataset/roadsign_voc
anno_path: valid.txt
label_list: label_list.txt
data_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']
TestDataset:
name: ImageFolder
anno_path: dataset/roadsign_voc/label_list.txt
configs/datasets/visdrone_detection.yml 0 → 100644
View file @ fccfdfa5
metric: COCO
num_classes: 10
TrainDataset:
!COCODataSet
image_dir: VisDrone2019-DET-train
anno_path: train.json
dataset_dir: dataset/visdrone
data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
EvalDataset:
!COCODataSet
image_dir: VisDrone2019-DET-val
anno_path: val.json
# image_dir: test_dev
# anno_path: test_dev.json
dataset_dir: dataset/visdrone
TestDataset:
!ImageFolder
anno_path: val.json
dataset_dir: dataset/visdrone
configs/datasets/voc.yml 0 → 100644
View file @ fccfdfa5
metric: VOC
map_type: 11point
num_classes: 20
TrainDataset:
name: VOCDataSet
dataset_dir: dataset/voc
anno_path: trainval.txt
label_list: label_list.txt
data_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']
EvalDataset:
name: VOCDataSet
dataset_dir: dataset/voc
anno_path: test.txt
label_list: label_list.txt
data_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']
TestDataset:
name: ImageFolder
anno_path: dataset/voc/label_list.txt
configs/runtime.yml 0 → 100644
View file @ fccfdfa5
use_gpu: true
use_xpu: false
use_mlu: false
use_npu: false
log_iter: 20
save_dir: output
snapshot_epoch: 1
print_flops: false
print_params: false
# Exporting the model
export:
post_process: True # Whether post-processing is included in the network when export model.
nms: True # Whether NMS is included in the network when export model.
benchmark: False # It is used to testing model performance, if set `True`, post-process and NMS will not be exported.
fuse_conv_bn: False
configs/yolov5/README.md 0 → 100644
View file @ fccfdfa5
# YOLOv5
## 内容
- [模型库](#模型库)
- [使用说明](#使用说明)
- [速度测试](#速度测试)
## 模型库
### 基础模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5-n | 640 | 16 | 300e | 1.5 | 28.0 | 45.7 | 1.87 | 4.52 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_n_300e_coco.pdparams) | [配置文件](./yolov5_n_300e_coco.yml) |
| YOLOv5-s | 640 | 16 | 300e | 2.6 | 37.6 | 56.7 | 7.24 | 16.54 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams) | [配置文件](./yolov5_s_300e_coco.yml) |
| YOLOv5-m | 640 | 16 | 300e | 5.2 | 45.4 | 64.1 | 21.19 | 49.08 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_m_300e_coco.pdparams) | [配置文件](./yolov5_m_300e_coco.yml) |
| YOLOv5-l | 640 | 16 | 300e | 7.9 | 48.9 | 67.1 | 46.56 | 109.32 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_l_300e_coco.pdparams) | [配置文件](./yolov5_l_300e_coco.yml) |
| YOLOv5-x | 640 | 16 | 300e | 13.7 | 50.6 | 68.7 | 86.75 | 205.92 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_x_300e_coco.pdparams) | [配置文件](./yolov5_x_300e_coco.yml) |
| YOLOv5-s ConvNeXt| 640 | 8 | 36e | - | 42.4 | 65.3 | 34.54 | 17.96 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_convnext_s_36e_coco.pdparams) | [配置文件](../convnext/yolov5_convnext_s_36e_coco.yml) |
### SSOD预训练模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5-s | 640 | 16 | 80e | 2.6 | 38.8 | 58.2 | 7.24 | 16.54 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_s_80e_ssod_finetune_coco.pdparams) | [配置文件](./yolov5_s_80e_ssod_finetune_coco.yml) |
### Objects-365预训练模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5-l | 640 | 16 | 30e | 7.9 | 49.3 | 67.8 | 46.56 | 109.32 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5_l_30e_obj365_finetune_coco.pdparams) | [配置文件](./yolov5_l_30e_obj365_finetune_coco.yml) |
### P6大尺度模型
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5p6-n | 1280 | 16 | 300e | - | 35.9 | 54.2 | 3.25 | 9.23 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_n_300e_coco.pdparams) | [配置文件](./yolov5p6_n_300e_coco.yml) |
| YOLOv5p6-s | 1280 | 16 | 300e | - | 44.5 | 63.3 | 12.63 | 33.81 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_s_300e_coco.pdparams) | [配置文件](./yolov5p6_s_300e_coco.yml) |
| YOLOv5p6-m | 1280 | 16 | 300e | - | 51.1 | 69.0 | 35.73 | 100.21 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_m_300e_coco.pdparams) | [配置文件](./yolov5p6_m_300e_coco.yml) |
| YOLOv5p6-l | 1280 | 8 | 300e | - | 53.4 | 71.0 | 76.77 | 223.09 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_l_300e_coco.pdparams) | [配置文件](./yolov5p6_l_300e_coco.yml) |
| YOLOv5p6-x | 1280 | 8 | 300e | - | 54.7 | 72.4 | 140.80 | 420.03 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5p6_x_300e_coco.pdparams) | [配置文件](./yolov5p6_x_300e_coco.yml) |
### [YOLOv5u](../yolov5u)
| 网络网络 | 输入尺寸 | 图片数/GPU | 学习率策略 | 模型推理耗时(ms) | mAP<sup>val<br>0.5:0.95 | mAP<sup>val<br>0.5 | Params(M) | FLOPs(G) | 下载链接 | 配置文件 |
| :------------- | :------- | :-------: | :------: | :------------: | :---------------------: | :----------------: |:---------: | :------: |:---------------: |:-----: |
| YOLOv5u-n | 640 | 16 | 300e | 1.61 | 34.5 | 49.7 | 2.65 | 7.79 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_n_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_n_300e_coco.yml) |
| YOLOv5u-s | 640 | 16 | 300e | 2.66 | 43.0 | 59.7 | 9.15 | 24.12 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_s_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_s_300e_coco.yml) |
| YOLOv5u-m | 640 | 16 | 300e | 5.50 | 49.0 | 65.7 | 25.11 | 64.42 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_m_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_m_300e_coco.yml) |
| YOLOv5u-l | 640 | 16 | 300e | 8.73 | 52.2 | 69.0 | 53.23 | 135.34 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_l_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_l_300e_coco.yml) |
| YOLOv5u-x | 640 | 16 | 300e | 15.49 | 53.1 | 69.9 | 97.28 | 246.89 | [下载链接](https://paddledet.bj.bcebos.com/models/yolov5u_x_300e_coco.pdparams) | [配置文件](./yolov5u/yolov5u_x_300e_coco.yml) |
**注意:**
- YOLOv5模型训练使用COCO train2017作为训练集，Box AP为在COCO val2017上的`mAP(IoU=0.5:0.95)`结果；
- YOLOv5u 模型表示YOLOv5结构使用YOLOv8的head和loss，是Anchor Free的检测方案，具体可参照[YOLOv5u](../yolov5u)；
- YOLOv5模型训练过程中默认使用8 GPUs进行混合精度训练，默认lr为0.01为8卡总batch_size的设置，如果**GPU卡数**或者每卡**batch size**发生改动，也不需要改动学习率，但为了保证高精度最好使用**总batch size大于64**的配置去训练；
- 模型推理耗时(ms)为TensorRT-FP16下测试的耗时，不包含数据预处理和模型输出后处理(NMS)的耗时。测试采用单卡Tesla T4 GPU，batch size=1，测试环境为**paddlepaddle-2.3.2**, **CUDA 11.2**, **CUDNN 8.2**, **GCC-8.2**, **TensorRT 8.0.3.4**，具体请参考[速度测试](#速度测试)。
- 如果你设置了`--run_benchmark=True`, 你首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
### 部署模型
| 网络模型 | 输入尺寸 | 导出后的权重(w/o NMS) | ONNX(w/o NMS) |
| :-------- | :--------: | :---------------------: | :----------------: |
| YOLOv5-n | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_n_300e_coco_wo_nms.onnx) |
| YOLOv5-s | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_s_300e_coco_wo_nms.onnx) |
| YOLOv5-m | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_m_300e_coco_wo_nms.onnx) |
| YOLOv5-l | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_l_300e_coco_wo_nms.onnx) |
| YOLOv5-x | 640 | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_w_nms.zip) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_wo_nms.zip) | [( w/ nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_w_nms.onnx) &#124; [( w/o nms)](https://paddledet.bj.bcebos.com/deploy/yoloseries/yolov5/yolov5_x_300e_coco_wo_nms.onnx) |
## 使用教程
### 0. **一键运行全流程**
将以下命令写在一个脚本文件里如```run.sh```，一键运行命令为：```sh run.sh```，也可命令行一句句去运行。
```bash
model_name=yolov5 # 可修改，如 ppyoloe
job_name=yolov5_s_300e_coco # 可修改，如 ppyoloe_plus_crn_s_80e_coco
config=configs/${model_name}/${job_name}.yml
log_dir=log_dir/${job_name}
# weights=https://bj.bcebos.com/v1/paddledet/models/${job_name}.pdparams
weights=output/${job_name}/model_final.pdparams
# 1.训练（单卡/多卡），加 --eval 表示边训边评估，加 --amp 表示混合精度训练
# CUDA_VISIBLE_DEVICES=0 python tools/train.py -c ${config} --eval --amp
python -m paddle.distributed.launch --log_dir=${log_dir} --gpus 0,1,2,3,4,5,6,7 tools/train.py -c ${config} --eval --amp
# 2.评估，加 --classwise 表示输出每一类mAP
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c ${config} -o weights=${weights} --classwise
# 3.预测 (单张图/图片文件夹）
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_img=demo/000000014439_640x640.jpg --draw_threshold=0.5
# CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c ${config} -o weights=${weights} --infer_dir=demo/ --draw_threshold=0.5
# 4.导出模型，以下3种模式选一种
## 普通导出，加trt表示用于trt加速，对NMS和silu激活函数提速明显
CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} # trt=True
## exclude_post_process去除后处理导出，返回和YOLOv5导出ONNX时相同格式的concat后的1个Tensor，是未缩放回原图的坐标+分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_post_process=True # trt=True
## exclude_nms去除NMS导出，返回2个Tensor，是缩放回原图后的坐标和分类置信度
# CUDA_VISIBLE_DEVICES=0 python tools/export_model.py -c ${config} -o weights=${weights} exclude_nms=True # trt=True
# 5.部署预测，注意不能使用 去除后处理 或 去除NMS 导出后的模型去预测
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU
# 6.部署测速，加 “--run_mode=trt_fp16” 表示在TensorRT FP16模式下测速，注意如需用到 trt_fp16 则必须为加 trt=True 导出的模型
CUDA_VISIBLE_DEVICES=0 python deploy/python/infer.py --model_dir=output_inference/${job_name} --image_file=demo/000000014439_640x640.jpg --device=GPU --run_benchmark=True # --run_mode=trt_fp16
# 7.onnx导出，一般结合 exclude_post_process去除后处理导出的模型
paddle2onnx --model_dir output_inference/${job_name} --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 12 --save_file ${job_name}.onnx
# 8.onnx trt测速
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp16
/usr/local/TensorRT-8.0.3.4/bin/trtexec --onnx=${job_name}.onnx --workspace=4096 --avgRuns=10 --shapes=input:1x3x640x640 --fp32
```
### 1. 训练
执行以下指令使用混合精度训练YOLOv5
```bash
python -m paddle.distributed.launch --gpus 0,1,2,3,4,5,6,7 tools/train.py -c configs/yolov5/yolov5_s_300e_coco.yml --amp --eval
```
**注意:**
- `--amp`表示开启混合精度训练以避免显存溢出，`--eval`表示边训边验证。
### 2. 评估
执行以下命令在单个GPU上评估COCO val2017数据集
```bash
CUDA_VISIBLE_DEVICES=0 python tools/eval.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams
```
### 3. 推理
使用以下命令在单张GPU上预测图片，使用`--infer_img`推理单张图片以及使用`--infer_dir`推理文件中的所有图片。
```bash
# 推理单张图片
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams --infer_img=demo/000000014439_640x640.jpg
# 推理文件中的所有图片
CUDA_VISIBLE_DEVICES=0 python tools/infer.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams --infer_dir=demo
```
### 4.导出模型
YOLOv5在GPU上推理部署或benchmark测速等需要通过`tools/export_model.py`导出模型。
当你**使用Paddle Inference但不使用TensorRT**时，运行以下的命令导出模型
```bash
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams
```
当你**使用Paddle Inference且使用TensorRT**时，需要指定`-o trt=True`来导出模型。
```bash
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams trt=True
```
如果你想将YOLOv5模型导出为**ONNX格式**，参考
[PaddleDetection模型导出为ONNX格式教程](../../deploy/EXPORT_ONNX_MODEL.md)，运行以下命令：
```bash
# 导出推理模型
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml --output_dir=output_inference -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams
# 安装paddle2onnx
pip install paddle2onnx
# 转换成onnx格式
paddle2onnx --model_dir output_inference/yolov5_s_300e_coco --model_filename model.pdmodel --params_filename model.pdiparams --opset_version 11 --save_file yolov5_s_300e_coco.onnx
```
**注意：** ONNX模型目前只支持batch_size=1
### 5.推理部署
YOLOv5可以使用以下方式进行部署：
- Paddle Inference [Python](../../deploy/python) & [C++](../../deploy/cpp)
- [Paddle-TensorRT](../../deploy/TENSOR_RT.md)
- [PaddleServing](https://github.com/PaddlePaddle/Serving)
- [PaddleSlim模型量化](../slim)
运行以下命令导出模型
```bash
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams trt=True
```
**注意：**
- trt=True表示**使用Paddle Inference且使用TensorRT**进行测速，速度会更快，默认不加即为False，表示**使用Paddle Inference但不使用TensorRT**进行测速。
- 如果是使用Paddle Inference在TensorRT FP16模式下部署，需要参考[Paddle Inference文档](https://www.paddlepaddle.org.cn/inference/master/user_guides/download_lib.html#python)，下载并安装与你的CUDA, CUDNN和TensorRT相应的wheel包。
#### 5.1.Python部署
`deploy/python/infer.py`使用上述导出后的Paddle Inference模型用于推理和benchnark测速，如果设置了`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
```bash
# Python部署推理单张图片
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu
# 推理文件夹下的所有图片
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_dir=demo/ --device=gpu
```
#### 5.2. C++部署
`deploy/cpp/build/main`使用上述导出后的Paddle Inference模型用于C++推理部署, 首先按照[docs](../../deploy/cpp/docs)编译安装环境。
```bash
# C++部署推理单张图片
./deploy/cpp/build/main --model_dir=output_inference/yolov5_s_300e_coco/ --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=GPU --threshold=0.5 --output_dir=cpp_infer_output/yolov5_s_300e_coco
```
## 速度测试
为了公平起见，在[模型库](#模型库)中的速度测试结果均为不包含数据预处理和模型输出后处理(NMS)的数据(与[YOLOv4(AlexyAB)](https://github.com/AlexeyAB/darknet)测试方法一致)，需要在导出模型时指定`-o exclude_nms=True`。测速需设置`--run_benchmark=True`, 首先需要安装以下依赖`pip install pynvml psutil GPUtil`。
**使用Paddle Inference但不使用TensorRT**进行测速，执行以下命令：
```bash
# 导出模型
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams exclude_nms=True
# 速度测试，使用run_benchmark=True
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --run_mode=paddle --device=gpu --run_benchmark=True
```
**使用Paddle Inference且使用TensorRT**进行测速，执行以下命令：
```bash
# 导出模型，使用trt=True
python tools/export_model.py -c configs/yolov5/yolov5_s_300e_coco.yml -o weights=https://paddledet.bj.bcebos.com/models/yolov5_s_300e_coco.pdparams exclude_nms=True trt=True
# 速度测试，使用run_benchmark=True
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True
# tensorRT-FP32测速
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp32
# tensorRT-FP16测速
python deploy/python/infer.py --model_dir=output_inference/yolov5_s_300e_coco --image_file=demo/000000014439_640x640.jpg --device=gpu --run_benchmark=True --run_mode=trt_fp16
```
**注意:**
- 导出模型时指定`-o exclude_nms=True`仅作为测速时用，这样导出的模型其推理部署预测的结果不是最终检出框的结果。
- [模型库](#模型库)中的速度测试结果为tensorRT-FP16测速后的最快速度，为不包含数据预处理和模型输出后处理(NMS)的耗时。
configs/yolov5/_base_/optimizer_300e.yml 0 → 100644
View file @ fccfdfa5
epoch: 300
LearningRate:
base_lr: 0.01
schedulers:
- !YOLOv5LRDecay
max_epochs: 300
min_lr_ratio: 0.01
- !ExpWarmup
epochs: 3
OptimizerBuilder:
optimizer:
type: Momentum
momentum: 0.937
use_nesterov: True
regularizer:
factor: 0.0005
type: L2
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment