Skip to content

GitLab

Commit c056df78 authored by sunxx1's avatar sunxx1
Browse files

Merge branch 'hepj-test' into 'main' 

增加VIT模型代码

See merge request dcutoolkit/deeplearing/dlexamples_new!39
parents 8ddf66c6 ddd93c0e
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 2858 additions and 0 deletions
PyTorch/NLP/Vision_Transformer/CODE_OF_CONDUCT.md 0 → 100644
View file @ c056df78
# Code of Conduct
## Our Pledge
In the interest of fostering an open and welcoming environment, we as
contributors and maintainers pledge to make participation in our project and
our community a harassment-free experience for everyone, regardless of age, body
size, disability, ethnicity, sex characteristics, gender identity and expression,
level of experience, education, socio-economic status, nationality, personal
appearance, race, religion, or sexual identity and orientation.
## Our Standards
Examples of behavior that contributes to creating a positive environment
include:
* Using welcoming and inclusive language
* Being respectful of differing viewpoints and experiences
* Gracefully accepting constructive criticism
* Focusing on what is best for the community
* Showing empathy towards other community members
Examples of unacceptable behavior by participants include:
* The use of sexualized language or imagery and unwelcome sexual attention or
advances
* Trolling, insulting/derogatory comments, and personal or political attacks
* Public or private harassment
* Publishing others' private information, such as a physical or electronic
address, without explicit permission
* Other conduct which could reasonably be considered inappropriate in a
professional setting
## Our Responsibilities
Project maintainers are responsible for clarifying the standards of acceptable
behavior and are expected to take appropriate and fair corrective action in
response to any instances of unacceptable behavior.
Project maintainers have the right and responsibility to remove, edit, or
reject comments, commits, code, wiki edits, issues, and other contributions
that are not aligned to this Code of Conduct, or to ban temporarily or
permanently any contributor for other behaviors that they deem inappropriate,
threatening, offensive, or harmful.
## Scope
This Code of Conduct applies within all project spaces, and it also applies when
an individual is representing the project or its community in public spaces.
Examples of representing a project or community include using an official
project e-mail address, posting via an official social media account, or acting
as an appointed representative at an online or offline event. Representation of
a project may be further defined and clarified by project maintainers.
This Code of Conduct also applies outside the project spaces when there is a
reasonable belief that an individual's behavior may have a negative impact on
the project or its community.
## Enforcement
Instances of abusive, harassing, or otherwise unacceptable behavior may be
reported by contacting the project team at <opensource-conduct@fb.com>. All
complaints will be reviewed and investigated and will result in a response that
is deemed necessary and appropriate to the circumstances. The project team is
obligated to maintain confidentiality with regard to the reporter of an incident.
Further details of specific enforcement policies may be posted separately.
Project maintainers who do not follow or enforce the Code of Conduct in good
faith may face temporary or permanent repercussions as determined by other
members of the project's leadership.
## Attribution
This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
[homepage]: https://www.contributor-covenant.org
For answers to common questions about this code of conduct, see
https://www.contributor-covenant.org/faq
PyTorch/NLP/Vision_Transformer/CONTRIBUTING.md 0 → 100644
View file @ c056df78
# Contributing to mae
We want to make contributing to this project as easy and transparent as
possible.
## Pull Requests
We actively welcome your pull requests.
1. Fork the repo and create your branch from `main`.
2. If you've added code that should be tested, add tests.
3. If you've changed APIs, update the documentation.
4. Ensure the test suite passes.
5. Make sure your code lints.
6. If you haven't already, complete the Contributor License Agreement ("CLA").
## Contributor License Agreement ("CLA")
In order to accept your pull request, we need you to submit a CLA. You only need
to do this once to work on any of Facebook's open source projects.
Complete your CLA here: <https://code.facebook.com/cla>
## Issues
We use GitHub issues to track public bugs. Please ensure your description is
clear and has sufficient instructions to be able to reproduce the issue.
Facebook has a [bounty program](https://www.facebook.com/whitehat/) for the safe
disclosure of security bugs. In those cases, please go through the process
outlined on that page and do not file a public issue.
## License
By contributing to mae, you agree that your contributions will be licensed
under the LICENSE file in the root directory of this source tree.
\ No newline at end of file
PyTorch/NLP/Vision_Transformer/FINETUNE.md 0 → 100644
View file @ c056df78
## Fine-tuning Pre-trained MAE for Classification
### Evaluation
As a sanity check, run evaluation using our ImageNet **fine-tuned** models:
<table><tbody>
<!-- START TABLE -->
<!-- TABLE HEADER -->
<th valign="bottom"></th>
<th valign="bottom">ViT-Base</th>
<th valign="bottom">ViT-Large</th>
<th valign="bottom">ViT-Huge</th>
<!-- TABLE BODY -->
<tr><td align="left">fine-tuned checkpoint</td>
<td align="center"><a href="https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_base.pth">download</a></td>
<td align="center"><a href="https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_large.pth">download</a></td>
<td align="center"><a href="https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_huge.pth">download</a></td>
</tr>
<tr><td align="left">md5</td>
<td align="center"><tt>1b25e9</tt></td>
<td align="center"><tt>51f550</tt></td>
<td align="center"><tt>2541f2</tt></td>
</tr>
<tr><td align="left">reference ImageNet accuracy</td>
<td align="center">83.664</td>
<td align="center">85.952</td>
<td align="center">86.928</td>
</tr>
</tbody></table>
Evaluate ViT-Base in a single GPU (`${IMAGENET_DIR}` is a directory containing `{train, val}` sets of ImageNet):
```
python main_finetune.py --eval --resume mae_finetuned_vit_base.pth --model vit_base_patch16 --batch_size 16 --data_path ${IMAGENET_DIR}
```
This should give:
```
* Acc@1 83.664 Acc@5 96.530 loss 0.731
```
Evaluate ViT-Large:
```
python main_finetune.py --eval --resume mae_finetuned_vit_large.pth --model vit_large_patch16 --batch_size 16 --data_path ${IMAGENET_DIR}
```
This should give:
```
* Acc@1 85.952 Acc@5 97.570 loss 0.646
```
Evaluate ViT-Huge:
```
python main_finetune.py --eval --resume mae_finetuned_vit_huge.pth --model vit_huge_patch14 --batch_size 16 --data_path ${IMAGENET_DIR}
```
This should give:
```
* Acc@1 86.928 Acc@5 98.088 loss 0.584
```
### Fine-tuning
Get our pre-trained checkpoints from [here](https://github.com/fairinternal/mae/#pre-trained-checkpoints).
To fine-tune with **multi-node distributed training**, run the following on 4 nodes with 8 GPUs each:
```
python submitit_finetune.py \
--job_dir ${JOB_DIR} \
--nodes 4 \
--batch_size 32 \
--model vit_base_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 100 \
--blr 5e-4 --layer_decay 0.65 \
--weight_decay 0.05 --drop_path 0.1 --reprob 0.25 --mixup 0.8 --cutmix 1.0 \
--dist_eval --data_path ${IMAGENET_DIR}
```
- Install submitit (`pip install submitit`) first.
- Here the effective batch size is 32 (`batch_size` per gpu) * 4 (`nodes`) * 8 (gpus per node) = 1024.
- `blr` is the base learning rate. The actual `lr` is computed by the [linear scaling rule](https://arxiv.org/abs/1706.02677): `lr` = `blr` * effective batch size / 256.
- We have run 4 trials with different random seeds. The resutls are 83.63, 83.66, 83.52, 83.46 (mean 83.57 and std 0.08).
- Training time is ~7h11m in 32 V100 GPUs.
Script for ViT-Large:
```
python submitit_finetune.py \
--job_dir ${JOB_DIR} \
--nodes 4 --use_volta32 \
--batch_size 32 \
--model vit_large_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 50 \
--blr 1e-3 --layer_decay 0.75 \
--weight_decay 0.05 --drop_path 0.2 --reprob 0.25 --mixup 0.8 --cutmix 1.0 \
--dist_eval --data_path ${IMAGENET_DIR}
```
- We have run 4 trials with different random seeds. The resutls are 85.95, 85.87, 85.76, 85.88 (mean 85.87 and std 0.07).
- Training time is ~8h52m in 32 V100 GPUs.
Script for ViT-Huge:
```
python submitit_finetune.py \
--job_dir ${JOB_DIR} \
--nodes 8 --use_volta32 \
--batch_size 16 \
--model vit_huge_patch14 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 50 \
--blr 1e-3 --layer_decay 0.75 \
--weight_decay 0.05 --drop_path 0.3 --reprob 0.25 --mixup 0.8 --cutmix 1.0 \
--dist_eval --data_path ${IMAGENET_DIR}
```
- Training time is ~13h9m in 64 V100 GPUs.
To fine-tune our pre-trained ViT-Base with **single-node training**, run the following on 1 node with 8 GPUs:
```
OMP_NUM_THREADS=1 python -m torch.distributed.launch --nproc_per_node=8 main_finetune.py \
--accum_iter 4 \
--batch_size 32 \
--model vit_base_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 100 \
--blr 5e-4 --layer_decay 0.65 \
--weight_decay 0.05 --drop_path 0.1 --mixup 0.8 --cutmix 1.0 --reprob 0.25 \
--dist_eval --data_path ${IMAGENET_DIR}
```
- Here the effective batch size is 32 (`batch_size` per gpu) * 4 (`accum_iter`) * 8 (gpus) = 1024. `--accum_iter 4` simulates 4 nodes.
#### Notes
- The [pre-trained models we provide](https://github.com/fairinternal/mae/#pre-trained-checkpoints) are trained with *normalized* pixels `--norm_pix_loss` (1600 epochs, Table 3 in paper). The fine-tuning hyper-parameters are slightly different from the default baseline using *unnormalized* pixels.
- The original MAE implementation was in TensorFlow+TPU with no explicit mixed precision. This re-implementation is in PyTorch+GPU with automatic mixed precision (`torch.cuda.amp`). We have observed different numerical behavior between the two platforms. In this repo, we use `--global_pool` for fine-tuning; using `--cls_token` performs similarly, but there is a chance of producing NaN when fine-tuning ViT-Huge in GPUs. We did not observe this issue in TPUs. Turning off amp could solve this issue, but is slower.
- Here we use RandErase following DeiT: `--reprob 0.25`. Its effect is smaller than random variance.
### Linear Probing
Run the following on 4 nodes with 8 GPUs each:
```
python submitit_linprobe.py \
--job_dir ${JOB_DIR} \
--nodes 4 \
--batch_size 512 \
--model vit_base_patch16 --cls_token \
--finetune ${PRETRAIN_CHKPT} \
--epochs 90 \
--blr 0.1 \
--weight_decay 0.0 \
--dist_eval --data_path ${IMAGENET_DIR}
```
- Here the effective batch size is 512 (`batch_size` per gpu) * 4 (`nodes`) * 8 (gpus per node) = 16384.
- `blr` is the base learning rate. The actual `lr` is computed by the [linear scaling rule](https://arxiv.org/abs/1706.02677): `lr` = `blr` * effective batch size / 256.
- Training time is ~2h20m for 90 epochs in 32 V100 GPUs.
- To run single-node training, follow the instruction in fine-tuning.
To train ViT-Large or ViT-Huge, set `--model vit_large_patch16` or `--model vit_huge_patch14`. It is sufficient to train 50 epochs `--epochs 50`.
This PT/GPU code produces *better* results for ViT-L/H (see the table below). This is likely caused by the system difference between TF and PT.
<table><tbody>
<!-- START TABLE -->
<!-- TABLE HEADER -->
<th valign="bottom"></th>
<th valign="bottom">ViT-Base</th>
<th valign="bottom">ViT-Large</th>
<th valign="bottom">ViT-Huge</th>
<!-- TABLE BODY -->
<tr><td align="left">paper (TF/TPU)</td>
<td align="center">68.0</td>
<td align="center">75.8</td>
<td align="center">76.6</td>
</tr>
<tr><td align="left">this repo (PT/GPU)</td>
<td align="center">67.8</td>
<td align="center">76.0</td>
<td align="center">77.2</td>
</tr>
</tbody></table>
PyTorch/NLP/Vision_Transformer/LICENSE 0 → 100644
View file @ c056df78
Attribution-NonCommercial 4.0 International
=======================================================================
Creative Commons Corporation ("Creative Commons") is not a law firm and
does not provide legal services or legal advice. Distribution of
Creative Commons public licenses does not create a lawyer-client or
other relationship. Creative Commons makes its licenses and related
information available on an "as-is" basis. Creative Commons gives no
warranties regarding its licenses, any material licensed under their
terms and conditions, or any related information. Creative Commons
disclaims all liability for damages resulting from their use to the
fullest extent possible.
Using Creative Commons Public Licenses
Creative Commons public licenses provide a standard set of terms and
conditions that creators and other rights holders may use to share
original works of authorship and other material subject to copyright
and certain other rights specified in the public license below. The
following considerations are for informational purposes only, are not
exhaustive, and do not form part of our licenses.
Considerations for licensors: Our public licenses are
intended for use by those authorized to give the public
permission to use material in ways otherwise restricted by
copyright and certain other rights. Our licenses are
irrevocable. Licensors should read and understand the terms
and conditions of the license they choose before applying it.
Licensors should also secure all rights necessary before
applying our licenses so that the public can reuse the
material as expected. Licensors should clearly mark any
material not subject to the license. This includes other CC-
licensed material, or material used under an exception or
limitation to copyright. More considerations for licensors:
wiki.creativecommons.org/Considerations_for_licensors
Considerations for the public: By using one of our public
licenses, a licensor grants the public permission to use the
licensed material under specified terms and conditions. If
the licensor's permission is not necessary for any reason--for
example, because of any applicable exception or limitation to
copyright--then that use is not regulated by the license. Our
licenses grant only permissions under copyright and certain
other rights that a licensor has authority to grant. Use of
the licensed material may still be restricted for other
reasons, including because others have copyright or other
rights in the material. A licensor may make special requests,
such as asking that all changes be marked or described.
Although not required by our licenses, you are encouraged to
respect those requests where reasonable. More_considerations
for the public:
wiki.creativecommons.org/Considerations_for_licensees
=======================================================================
Creative Commons Attribution-NonCommercial 4.0 International Public
License
By exercising the Licensed Rights (defined below), You accept and agree
to be bound by the terms and conditions of this Creative Commons
Attribution-NonCommercial 4.0 International Public License ("Public
License"). To the extent this Public License may be interpreted as a
contract, You are granted the Licensed Rights in consideration of Your
acceptance of these terms and conditions, and the Licensor grants You
such rights in consideration of benefits the Licensor receives from
making the Licensed Material available under these terms and
conditions.
Section 1 -- Definitions.
a. Adapted Material means material subject to Copyright and Similar
Rights that is derived from or based upon the Licensed Material
and in which the Licensed Material is translated, altered,
arranged, transformed, or otherwise modified in a manner requiring
permission under the Copyright and Similar Rights held by the
Licensor. For purposes of this Public License, where the Licensed
Material is a musical work, performance, or sound recording,
Adapted Material is always produced where the Licensed Material is
synched in timed relation with a moving image.
b. Adapter's License means the license You apply to Your Copyright
and Similar Rights in Your contributions to Adapted Material in
accordance with the terms and conditions of this Public License.
c. Copyright and Similar Rights means copyright and/or similar rights
closely related to copyright including, without limitation,
performance, broadcast, sound recording, and Sui Generis Database
Rights, without regard to how the rights are labeled or
categorized. For purposes of this Public License, the rights
specified in Section 2(b)(1)-(2) are not Copyright and Similar
Rights.
d. Effective Technological Measures means those measures that, in the
absence of proper authority, may not be circumvented under laws
fulfilling obligations under Article 11 of the WIPO Copyright
Treaty adopted on December 20, 1996, and/or similar international
agreements.
e. Exceptions and Limitations means fair use, fair dealing, and/or
any other exception or limitation to Copyright and Similar Rights
that applies to Your use of the Licensed Material.
f. Licensed Material means the artistic or literary work, database,
or other material to which the Licensor applied this Public
License.
g. Licensed Rights means the rights granted to You subject to the
terms and conditions of this Public License, which are limited to
all Copyright and Similar Rights that apply to Your use of the
Licensed Material and that the Licensor has authority to license.
h. Licensor means the individual(s) or entity(ies) granting rights
under this Public License.
i. NonCommercial means not primarily intended for or directed towards
commercial advantage or monetary compensation. For purposes of
this Public License, the exchange of the Licensed Material for
other material subject to Copyright and Similar Rights by digital
file-sharing or similar means is NonCommercial provided there is
no payment of monetary compensation in connection with the
exchange.
j. Share means to provide material to the public by any means or
process that requires permission under the Licensed Rights, such
as reproduction, public display, public performance, distribution,
dissemination, communication, or importation, and to make material
available to the public including in ways that members of the
public may access the material from a place and at a time
individually chosen by them.
k. Sui Generis Database Rights means rights other than copyright
resulting from Directive 96/9/EC of the European Parliament and of
the Council of 11 March 1996 on the legal protection of databases,
as amended and/or succeeded, as well as other essentially
equivalent rights anywhere in the world.
l. You means the individual or entity exercising the Licensed Rights
under this Public License. Your has a corresponding meaning.
Section 2 -- Scope.
a. License grant.
1. Subject to the terms and conditions of this Public License,
the Licensor hereby grants You a worldwide, royalty-free,
non-sublicensable, non-exclusive, irrevocable license to
exercise the Licensed Rights in the Licensed Material to:
a. reproduce and Share the Licensed Material, in whole or
in part, for NonCommercial purposes only; and
b. produce, reproduce, and Share Adapted Material for
NonCommercial purposes only.
2. Exceptions and Limitations. For the avoidance of doubt, where
Exceptions and Limitations apply to Your use, this Public
License does not apply, and You do not need to comply with
its terms and conditions.
3. Term. The term of this Public License is specified in Section
6(a).
4. Media and formats; technical modifications allowed. The
Licensor authorizes You to exercise the Licensed Rights in
all media and formats whether now known or hereafter created,
and to make technical modifications necessary to do so. The
Licensor waives and/or agrees not to assert any right or
authority to forbid You from making technical modifications
necessary to exercise the Licensed Rights, including
technical modifications necessary to circumvent Effective
Technological Measures. For purposes of this Public License,
simply making modifications authorized by this Section 2(a)
(4) never produces Adapted Material.
5. Downstream recipients.
a. Offer from the Licensor -- Licensed Material. Every
recipient of the Licensed Material automatically
receives an offer from the Licensor to exercise the
Licensed Rights under the terms and conditions of this
Public License.
b. No downstream restrictions. You may not offer or impose
any additional or different terms or conditions on, or
apply any Effective Technological Measures to, the
Licensed Material if doing so restricts exercise of the
Licensed Rights by any recipient of the Licensed
Material.
6. No endorsement. Nothing in this Public License constitutes or
may be construed as permission to assert or imply that You
are, or that Your use of the Licensed Material is, connected
with, or sponsored, endorsed, or granted official status by,
the Licensor or others designated to receive attribution as
provided in Section 3(a)(1)(A)(i).
b. Other rights.
1. Moral rights, such as the right of integrity, are not
licensed under this Public License, nor are publicity,
privacy, and/or other similar personality rights; however, to
the extent possible, the Licensor waives and/or agrees not to
assert any such rights held by the Licensor to the limited
extent necessary to allow You to exercise the Licensed
Rights, but not otherwise.
2. Patent and trademark rights are not licensed under this
Public License.
3. To the extent possible, the Licensor waives any right to
collect royalties from You for the exercise of the Licensed
Rights, whether directly or through a collecting society
under any voluntary or waivable statutory or compulsory
licensing scheme. In all other cases the Licensor expressly
reserves any right to collect such royalties, including when
the Licensed Material is used other than for NonCommercial
purposes.
Section 3 -- License Conditions.
Your exercise of the Licensed Rights is expressly made subject to the
following conditions.
a. Attribution.
1. If You Share the Licensed Material (including in modified
form), You must:
a. retain the following if it is supplied by the Licensor
with the Licensed Material:
i. identification of the creator(s) of the Licensed
Material and any others designated to receive
attribution, in any reasonable manner requested by
the Licensor (including by pseudonym if
designated);
ii. a copyright notice;
iii. a notice that refers to this Public License;
iv. a notice that refers to the disclaimer of
warranties;
v. a URI or hyperlink to the Licensed Material to the
extent reasonably practicable;
b. indicate if You modified the Licensed Material and
retain an indication of any previous modifications; and
c. indicate the Licensed Material is licensed under this
Public License, and include the text of, or the URI or
hyperlink to, this Public License.
2. You may satisfy the conditions in Section 3(a)(1) in any
reasonable manner based on the medium, means, and context in
which You Share the Licensed Material. For example, it may be
reasonable to satisfy the conditions by providing a URI or
hyperlink to a resource that includes the required
information.
3. If requested by the Licensor, You must remove any of the
information required by Section 3(a)(1)(A) to the extent
reasonably practicable.
4. If You Share Adapted Material You produce, the Adapter's
License You apply must not prevent recipients of the Adapted
Material from complying with this Public License.
Section 4 -- Sui Generis Database Rights.
Where the Licensed Rights include Sui Generis Database Rights that
apply to Your use of the Licensed Material:
a. for the avoidance of doubt, Section 2(a)(1) grants You the right
to extract, reuse, reproduce, and Share all or a substantial
portion of the contents of the database for NonCommercial purposes
only;
b. if You include all or a substantial portion of the database
contents in a database in which You have Sui Generis Database
Rights, then the database in which You have Sui Generis Database
Rights (but not its individual contents) is Adapted Material; and
c. You must comply with the conditions in Section 3(a) if You Share
all or a substantial portion of the contents of the database.
For the avoidance of doubt, this Section 4 supplements and does not
replace Your obligations under this Public License where the Licensed
Rights include other Copyright and Similar Rights.
Section 5 -- Disclaimer of Warranties and Limitation of Liability.
a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.
b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.
c. The disclaimer of warranties and limitation of liability provided
above shall be interpreted in a manner that, to the extent
possible, most closely approximates an absolute disclaimer and
waiver of all liability.
Section 6 -- Term and Termination.
a. This Public License applies for the term of the Copyright and
Similar Rights licensed here. However, if You fail to comply with
this Public License, then Your rights under this Public License
terminate automatically.
b. Where Your right to use the Licensed Material has terminated under
Section 6(a), it reinstates:
1. automatically as of the date the violation is cured, provided
it is cured within 30 days of Your discovery of the
violation; or
2. upon express reinstatement by the Licensor.
For the avoidance of doubt, this Section 6(b) does not affect any
right the Licensor may have to seek remedies for Your violations
of this Public License.
c. For the avoidance of doubt, the Licensor may also offer the
Licensed Material under separate terms or conditions or stop
distributing the Licensed Material at any time; however, doing so
will not terminate this Public License.
d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
License.
Section 7 -- Other Terms and Conditions.
a. The Licensor shall not be bound by any additional or different
terms or conditions communicated by You unless expressly agreed.
b. Any arrangements, understandings, or agreements regarding the
Licensed Material not stated herein are separate from and
independent of the terms and conditions of this Public License.
Section 8 -- Interpretation.
a. For the avoidance of doubt, this Public License does not, and
shall not be interpreted to, reduce, limit, restrict, or impose
conditions on any use of the Licensed Material that could lawfully
be made without permission under this Public License.
b. To the extent possible, if any provision of this Public License is
deemed unenforceable, it shall be automatically reformed to the
minimum extent necessary to make it enforceable. If the provision
cannot be reformed, it shall be severed from this Public License
without affecting the enforceability of the remaining terms and
conditions.
c. No term or condition of this Public License will be waived and no
failure to comply consented to unless expressly agreed to by the
Licensor.
d. Nothing in this Public License constitutes or may be interpreted
as a limitation upon, or waiver of, any privileges and immunities
that apply to the Licensor or You, including from the legal
processes of any jurisdiction or authority.
=======================================================================
Creative Commons is not a party to its public
licenses. Notwithstanding, Creative Commons may elect to apply one of
its public licenses to material it publishes and in those instances
will be considered the “Licensor.” The text of the Creative Commons
public licenses is dedicated to the public domain under the CC0 Public
Domain Dedication. Except for the limited purpose of indicating that
material is shared under a Creative Commons public license or as
otherwise permitted by the Creative Commons policies published at
creativecommons.org/policies, Creative Commons does not authorize the
use of the trademark "Creative Commons" or any other trademark or logo
of Creative Commons without its prior written consent including,
without limitation, in connection with any unauthorized modifications
to any of its public licenses or any other arrangements,
understandings, or agreements concerning use of licensed material. For
the avoidance of doubt, this paragraph does not form part of the
public licenses.
Creative Commons may be contacted at creativecommons.org.
\ No newline at end of file
PyTorch/NLP/Vision_Transformer/PRETRAIN.md 0 → 100644
View file @ c056df78
## Pre-training MAE
To pre-train ViT-Large (recommended default) with **multi-node distributed training**, run the following on 8 nodes with 8 GPUs each:
```
python submitit_pretrain.py \
--job_dir ${JOB_DIR} \
--nodes 8 \
--use_volta32 \
--batch_size 64 \
--model mae_vit_large_patch16 \
--norm_pix_loss \
--mask_ratio 0.75 \
--epochs 800 \
--warmup_epochs 40 \
--blr 1.5e-4 --weight_decay 0.05 \
--data_path ${IMAGENET_DIR}
```
- Here the effective batch size is 64 (`batch_size` per gpu) * 8 (`nodes`) * 8 (gpus per node) = 4096. If memory or # gpus is limited, use `--accum_iter` to maintain the effective batch size, which is `batch_size` (per gpu) * `nodes` * 8 (gpus per node) * `accum_iter`.
- `blr` is the base learning rate. The actual `lr` is computed by the [linear scaling rule](https://arxiv.org/abs/1706.02677): `lr` = `blr` * effective batch size / 256.
- Here we use `--norm_pix_loss` as the target for better representation learning. To train a baseline model (e.g., for visualization), use pixel-based construction and turn off `--norm_pix_loss`.
- The exact same hyper-parameters and configs (initialization, augmentation, etc.) are used as our TF/TPU implementation. In our sanity checks, this PT/GPU re-implementation can reproduce the TF/TPU results within reasonable random variation. We get 85.5% [fine-tuning](FINETUNE.md) accuracy by pre-training ViT-Large for 800 epochs (85.4% in paper Table 1d with TF/TPU).
- Training time is ~42h in 64 V100 GPUs (800 epochs).
To train ViT-Base or ViT-Huge, set `--model mae_vit_base_patch16` or `--model mae_vit_huge_patch14`.
PyTorch/NLP/Vision_Transformer/README-origin.md 0 → 100644
View file @ c056df78
## Masked Autoencoders: A PyTorch Implementation
<p align="center">
<img src="https://user-images.githubusercontent.com/11435359/146857310-f258c86c-fde6-48e8-9cee-badd2b21bd2c.png" width="480">
</p>
This is a PyTorch/GPU re-implementation of the paper [Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377):
```
@Article{MaskedAutoencoders2021,
author = {Kaiming He and Xinlei Chen and Saining Xie and Yanghao Li and Piotr Doll{\'a}r and Ross Girshick},
journal = {arXiv:2111.06377},
title = {Masked Autoencoders Are Scalable Vision Learners},
year = {2021},
}
```
* The original implementation was in TensorFlow+TPU. This re-implementation is in PyTorch+GPU.
* This repo is a modification on the [DeiT repo](https://github.com/facebookresearch/deit). Installation and preparation follow that repo.
* This repo is based on [`timm==0.3.2`](https://github.com/rwightman/pytorch-image-models), for which a [fix](https://github.com/rwightman/pytorch-image-models/issues/420#issuecomment-776459842) is needed to work with PyTorch 1.8.1+.
### Catalog
- [x] Visualization demo
- [x] Pre-trained checkpoints + fine-tuning code
- [x] Pre-training code
### Visualization demo
Run our interactive visualization demo using [Colab notebook](https://colab.research.google.com/github/facebookresearch/mae/blob/main/demo/mae_visualize.ipynb) (no GPU needed):
<p align="center">
<img src="https://user-images.githubusercontent.com/11435359/147859292-77341c70-2ed8-4703-b153-f505dcb6f2f8.png" width="600">
</p>
### Fine-tuning with pre-trained checkpoints
The following table provides the pre-trained checkpoints used in the paper, converted from TF/TPU to PT/GPU:
<table><tbody>
<!-- START TABLE -->
<!-- TABLE HEADER -->
<th valign="bottom"></th>
<th valign="bottom">ViT-Base</th>
<th valign="bottom">ViT-Large</th>
<th valign="bottom">ViT-Huge</th>
<!-- TABLE BODY -->
<tr><td align="left">pre-trained checkpoint</td>
<td align="center"><a href="https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_base.pth">download</a></td>
<td align="center"><a href="https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_large.pth">download</a></td>
<td align="center"><a href="https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_huge.pth">download</a></td>
</tr>
<tr><td align="left">md5</td>
<td align="center"><tt>8cad7c</tt></td>
<td align="center"><tt>b8b06e</tt></td>
<td align="center"><tt>9bdbb0</tt></td>
</tr>
</tbody></table>
The fine-tuning instruction is in [FINETUNE.md](FINETUNE.md).
By fine-tuning these pre-trained models, we rank #1 in these classification tasks (detailed in the paper):
<table><tbody>
<!-- START TABLE -->
<!-- TABLE HEADER -->
<th valign="bottom"></th>
<th valign="bottom">ViT-B</th>
<th valign="bottom">ViT-L</th>
<th valign="bottom">ViT-H</th>
<th valign="bottom">ViT-H<sub>448</sub></th>
<td valign="bottom" style="color:#C0C0C0">prev best</td>
<!-- TABLE BODY -->
<tr><td align="left">ImageNet-1K (no external data)</td>
<td align="center">83.6</td>
<td align="center">85.9</td>
<td align="center">86.9</td>
<td align="center"><b>87.8</b></td>
<td align="center" style="color:#C0C0C0">87.1</td>
</tr>
<td colspan="5"><font size="1"><em>following are evaluation of the same model weights (fine-tuned in original ImageNet-1K):</em></font></td>
<tr>
</tr>
<tr><td align="left">ImageNet-Corruption (error rate) </td>
<td align="center">51.7</td>
<td align="center">41.8</td>
<td align="center"><b>33.8</b></td>
<td align="center">36.8</td>
<td align="center" style="color:#C0C0C0">42.5</td>
</tr>
<tr><td align="left">ImageNet-Adversarial</td>
<td align="center">35.9</td>
<td align="center">57.1</td>
<td align="center">68.2</td>
<td align="center"><b>76.7</b></td>
<td align="center" style="color:#C0C0C0">35.8</td>
</tr>
<tr><td align="left">ImageNet-Rendition</td>
<td align="center">48.3</td>
<td align="center">59.9</td>
<td align="center">64.4</td>
<td align="center"><b>66.5</b></td>
<td align="center" style="color:#C0C0C0">48.7</td>
</tr>
<tr><td align="left">ImageNet-Sketch</td>
<td align="center">34.5</td>
<td align="center">45.3</td>
<td align="center">49.6</td>
<td align="center"><b>50.9</b></td>
<td align="center" style="color:#C0C0C0">36.0</td>
</tr>
<td colspan="5"><font size="1"><em>following are transfer learning by fine-tuning the pre-trained MAE on the target dataset:</em></font></td>
</tr>
<tr><td align="left">iNaturalists 2017</td>
<td align="center">70.5</td>
<td align="center">75.7</td>
<td align="center">79.3</td>
<td align="center"><b>83.4</b></td>
<td align="center" style="color:#C0C0C0">75.4</td>
</tr>
<tr><td align="left">iNaturalists 2018</td>
<td align="center">75.4</td>
<td align="center">80.1</td>
<td align="center">83.0</td>
<td align="center"><b>86.8</b></td>
<td align="center" style="color:#C0C0C0">81.2</td>
</tr>
<tr><td align="left">iNaturalists 2019</td>
<td align="center">80.5</td>
<td align="center">83.4</td>
<td align="center">85.7</td>
<td align="center"><b>88.3</b></td>
<td align="center" style="color:#C0C0C0">84.1</td>
</tr>
<tr><td align="left">Places205</td>
<td align="center">63.9</td>
<td align="center">65.8</td>
<td align="center">65.9</td>
<td align="center"><b>66.8</b></td>
<td align="center" style="color:#C0C0C0">66.0</td>
</tr>
<tr><td align="left">Places365</td>
<td align="center">57.9</td>
<td align="center">59.4</td>
<td align="center">59.8</td>
<td align="center"><b>60.3</b></td>
<td align="center" style="color:#C0C0C0">58.0</td>
</tr>
</tbody></table>
### Pre-training
The pre-training instruction is in [PRETRAIN.md](PRETRAIN.md).
### License
This project is under the CC-BY-NC 4.0 license. See [LICENSE](LICENSE) for details.
PyTorch/NLP/Vision_Transformer/README.md 0 → 100644
View file @ c056df78
# 简介
pytorch框架下VIT模型预训练和微调任务，使用数据集为pytorch版本的ImageNet数据集
[代码地址](https://github.com/facebookresearch/mae)
# 数据集&模型地址
```
#昆山
/public/software/apps/DeepLearning/Data/ImageNet-pytorch
#乌镇
/public/DL_DATA/ImageNet-pytorch
```
模型下载地址（finetune时加载）：
[mae_pretrain_vit_base.pth](https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_base.pth)
[mae_pretrain_vit_large.pth](https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_large.pth)
[mae_pretrain_vit_huge.pth](https://dl.fbaipublicfiles.com/mae/pretrain/mae_pretrain_vit_huge.pth)
# 运行前准备
## 依赖包版本
```
需要安装timm包，且版本必须为0.3.2，一些重要依赖包版本如下：
tensorboard 2.0.2
protobuf 3.20.0
matplotlib 3.5.3
numpy 1.21.6
absl-py 1.2.0
```
## 报错
```
报错ImportError: cannot import name ‘container_abcs‘ from ‘torch._six‘
这是由于在1.8版本之后container_abcs就已经被移除了，而代码是基于1.8之前别的版本，所以导入方式不同会出现这样的错误：cannot import name ‘container_abcs’ from ‘torch._six’
因此使用不同版本的torch会出现这个问题。
1.8以下版本使用 ：
from torch._six import container_abcs；
1.8以上版本使用（改成这个）：
import collections.abc as container_abcs
int_classes = int
string_classes = str
```
## 环境变量
```
#环境变量
IMAGENET_DIR 数据集地址
PRETRAIN_CHKPT 模型文件地址
```
# 预训练任务
```
#模型参数
model 模型类型,可选:mae_vit_base_patch16,mae_vit_large_patch16,mae_vit_huge_patch14
预训练不需要加载模型，只需给出模型类型
```
## 单卡
./run_pretrain.sh
```
#!/bin/bash
export HIP_VISIBLE_DEVICE=0
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
python3 main_pretrain.py --epochs 1 --model mae_vit_base_patch16 --batch_size 64 --model mae_vit_base_patch16 --data_path ${IMAGENET_DIR}
```
## 多卡
./run_pretrain-4.sh
```
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
export HIP_VISIBLE_DEVICE=0,1,2,3
OMP_NUM_THREADS=1 python3 -m torch.distributed.launch --nproc_per_node=4 main_pretrain.py --epochs 1 --accum_iter 4 --model mae_vit_base_patch16 --batch_size 64 --model mae_vit_base_patch16 --data_path ${IMAGENET_DIR}
```
# 微调任务
```
#模型参数
model 模型类型,可选:vit_base_patch16,vit_large_patch16,vit_huge_patch14,注意这个参数要与PRETRAIN_CHKPT所给模型地址相匹配。例如：vit_base_patch16对应mae_pretrain_vit_base.pth
```
## 单卡
./run-finetune.sh
```
export HIP_VISIBLE_DEVICE=0
export PRETRAIN_CHKPT=/work/home/hepj/model/VIT/mae_pretrain_vit_base.pth
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
python3 main_finetune.py \
--batch_size 32 \
--model vit_base_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 1 \
--blr 5e-4 --layer_decay 0.65 \
--weight_decay 0.05 --drop_path 0.1 --mixup 0.8 --cutmix 1.0 --reprob 0.25 \
--dist_eval --data_path ${IMAGENET_DIR}
```
## 多卡
./run-finetune-4.sh
```
export HIP_VISIBLE_DEVICE=0,1,2,3
export PRETRAIN_CHKPT=/work/home/hepj/model/VIT/mae_pretrain_vit_base.pth
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
OMP_NUM_THREADS=1 python3 -m torch.distributed.launch --nproc_per_node=4 main_finetune.py \
--accum_iter 4 \
--batch_size 32 \
--model vit_base_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 1 \
--blr 5e-4 --layer_decay 0.65 \
--weight_decay 0.05 --drop_path 0.1 --mixup 0.8 --cutmix 1.0 --reprob 0.25 \
--dist_eval --data_path ${IMAGENET_DIR}
```
# 结果验证
验证使用的模型为mae_finetuned_vit_xxx.pth，下载地址：
[mae_finetuned_vit_base.pth](https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_base.pth)
[mae_finetuned_vit_large.pth](https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_large.pth)
[mae_finetuned_vit_huge.pth](https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_huge.pth)
./run_eval.sh
```
#注意 model 与 resume模型对应
export PRETRAIN_CHKPT=/work/home/hepj/model/VIT/mae_finetuned_vit_base.pth
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
python main_finetune.py\
--eval\
--resume ${PRETRAIN_CHKPT}\
--model vit_base_patch16\
--batch_size 16\
--data_path ${IMAGENET_DIR}
```
PyTorch/NLP/Vision_Transformer/demo/mae_visualize.ipynb 0 → 100644
View file @ c056df78
This diff is collapsed.
PyTorch/NLP/Vision_Transformer/engine_finetune.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# BEiT: https://github.com/microsoft/unilm/tree/master/beit
# --------------------------------------------------------
import math
import sys
from typing import Iterable, Optional
import torch
from timm.data import Mixup
from timm.utils import accuracy
import util.misc as misc
import util.lr_sched as lr_sched
def train_one_epoch(model: torch.nn.Module, criterion: torch.nn.Module,
data_loader: Iterable, optimizer: torch.optim.Optimizer,
device: torch.device, epoch: int, loss_scaler, max_norm: float = 0,
mixup_fn: Optional[Mixup] = None, log_writer=None,
args=None):
model.train(True)
metric_logger = misc.MetricLogger(delimiter=" ")
metric_logger.add_meter('lr', misc.SmoothedValue(window_size=1, fmt='{value:.6f}'))
header = 'Epoch: [{}]'.format(epoch)
print_freq = 20
accum_iter = args.accum_iter
optimizer.zero_grad()
if log_writer is not None:
print('log_dir: {}'.format(log_writer.log_dir))
for data_iter_step, (samples, targets) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):
# we use a per iteration (instead of per epoch) lr scheduler
if data_iter_step % accum_iter == 0:
lr_sched.adjust_learning_rate(optimizer, data_iter_step / len(data_loader) + epoch, args)
samples = samples.to(device, non_blocking=True)
targets = targets.to(device, non_blocking=True)
if mixup_fn is not None:
samples, targets = mixup_fn(samples, targets)
with torch.cuda.amp.autocast():
outputs = model(samples)
loss = criterion(outputs, targets)
loss_value = loss.item()
if not math.isfinite(loss_value):
print("Loss is {}, stopping training".format(loss_value))
sys.exit(1)
loss /= accum_iter
loss_scaler(loss, optimizer, clip_grad=max_norm,
parameters=model.parameters(), create_graph=False,
update_grad=(data_iter_step + 1) % accum_iter == 0)
if (data_iter_step + 1) % accum_iter == 0:
optimizer.zero_grad()
torch.cuda.synchronize()
metric_logger.update(loss=loss_value)
min_lr = 10.
max_lr = 0.
for group in optimizer.param_groups:
min_lr = min(min_lr, group["lr"])
max_lr = max(max_lr, group["lr"])
metric_logger.update(lr=max_lr)
loss_value_reduce = misc.all_reduce_mean(loss_value)
if log_writer is not None and (data_iter_step + 1) % accum_iter == 0:
""" We use epoch_1000x as the x-axis in tensorboard.
This calibrates different curves when batch size changes.
"""
epoch_1000x = int((data_iter_step / len(data_loader) + epoch) * 1000)
log_writer.add_scalar('loss', loss_value_reduce, epoch_1000x)
log_writer.add_scalar('lr', max_lr, epoch_1000x)
# gather the stats from all processes
metric_logger.synchronize_between_processes()
print("Averaged stats:", metric_logger)
return {k: meter.global_avg for k, meter in metric_logger.meters.items()}
@torch.no_grad()
def evaluate(data_loader, model, device):
criterion = torch.nn.CrossEntropyLoss()
metric_logger = misc.MetricLogger(delimiter=" ")
header = 'Test:'
# switch to evaluation mode
model.eval()
for batch in metric_logger.log_every(data_loader, 10, header):
images = batch[0]
target = batch[-1]
images = images.to(device, non_blocking=True)
target = target.to(device, non_blocking=True)
# compute output
with torch.cuda.amp.autocast():
output = model(images)
loss = criterion(output, target)
acc1, acc5 = accuracy(output, target, topk=(1, 5))
batch_size = images.shape[0]
metric_logger.update(loss=loss.item())
metric_logger.meters['acc1'].update(acc1.item(), n=batch_size)
metric_logger.meters['acc5'].update(acc5.item(), n=batch_size)
# gather the stats from all processes
metric_logger.synchronize_between_processes()
print('* Acc@1 {top1.global_avg:.3f} Acc@5 {top5.global_avg:.3f} loss {losses.global_avg:.3f}'
.format(top1=metric_logger.acc1, top5=metric_logger.acc5, losses=metric_logger.loss))
return {k: meter.global_avg for k, meter in metric_logger.meters.items()}
\ No newline at end of file
PyTorch/NLP/Vision_Transformer/engine_pretrain.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# BEiT: https://github.com/microsoft/unilm/tree/master/beit
# --------------------------------------------------------
import math
import sys
from typing import Iterable
import torch
import util.misc as misc
import util.lr_sched as lr_sched
def train_one_epoch(model: torch.nn.Module,
data_loader: Iterable, optimizer: torch.optim.Optimizer,
device: torch.device, epoch: int, loss_scaler,
log_writer=None,
args=None):
model.train(True)
metric_logger = misc.MetricLogger(delimiter=" ")
metric_logger.add_meter('lr', misc.SmoothedValue(window_size=1, fmt='{value:.6f}'))
header = 'Epoch: [{}]'.format(epoch)
print_freq = 20
accum_iter = args.accum_iter
optimizer.zero_grad()
if log_writer is not None:
print('log_dir: {}'.format(log_writer.log_dir))
for data_iter_step, (samples, _) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):
# we use a per iteration (instead of per epoch) lr scheduler
if data_iter_step % accum_iter == 0:
lr_sched.adjust_learning_rate(optimizer, data_iter_step / len(data_loader) + epoch, args)
samples = samples.to(device, non_blocking=True)
with torch.cuda.amp.autocast():
loss, _, _ = model(samples, mask_ratio=args.mask_ratio)
loss_value = loss.item()
if not math.isfinite(loss_value):
print("Loss is {}, stopping training".format(loss_value))
sys.exit(1)
loss /= accum_iter
loss_scaler(loss, optimizer, parameters=model.parameters(),
update_grad=(data_iter_step + 1) % accum_iter == 0)
if (data_iter_step + 1) % accum_iter == 0:
optimizer.zero_grad()
torch.cuda.synchronize()
metric_logger.update(loss=loss_value)
lr = optimizer.param_groups[0]["lr"]
metric_logger.update(lr=lr)
loss_value_reduce = misc.all_reduce_mean(loss_value)
if log_writer is not None and (data_iter_step + 1) % accum_iter == 0:
""" We use epoch_1000x as the x-axis in tensorboard.
This calibrates different curves when batch size changes.
"""
epoch_1000x = int((data_iter_step / len(data_loader) + epoch) * 1000)
log_writer.add_scalar('train_loss', loss_value_reduce, epoch_1000x)
log_writer.add_scalar('lr', lr, epoch_1000x)
# gather the stats from all processes
metric_logger.synchronize_between_processes()
print("Averaged stats:", metric_logger)
return {k: meter.global_avg for k, meter in metric_logger.meters.items()}
\ No newline at end of file
PyTorch/NLP/Vision_Transformer/main_finetune.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# BEiT: https://github.com/microsoft/unilm/tree/master/beit
# --------------------------------------------------------
import argparse
import datetime
import json
import numpy as np
import os
import time
from pathlib import Path
import torch
import torch.backends.cudnn as cudnn
from torch.utils.tensorboard import SummaryWriter
import timm
assert timm.__version__ == "0.3.2" # version check
from timm.models.layers import trunc_normal_
from timm.data.mixup import Mixup
from timm.loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
import util.lr_decay as lrd
import util.misc as misc
from util.datasets import build_dataset
from util.pos_embed import interpolate_pos_embed
from util.misc import NativeScalerWithGradNormCount as NativeScaler
import models_vit
from engine_finetune import train_one_epoch, evaluate
def get_args_parser():
parser = argparse.ArgumentParser('MAE fine-tuning for image classification', add_help=False)
parser.add_argument('--batch_size', default=64, type=int,
help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
parser.add_argument('--epochs', default=50, type=int)
parser.add_argument('--accum_iter', default=1, type=int,
help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')
# Model parameters
parser.add_argument('--model', default='vit_large_patch16', type=str, metavar='MODEL',
help='Name of model to train')
parser.add_argument('--input_size', default=224, type=int,
help='images input size')
parser.add_argument('--drop_path', type=float, default=0.1, metavar='PCT',
help='Drop path rate (default: 0.1)')
# Optimizer parameters
parser.add_argument('--clip_grad', type=float, default=None, metavar='NORM',
help='Clip gradient norm (default: None, no clipping)')
parser.add_argument('--weight_decay', type=float, default=0.05,
help='weight decay (default: 0.05)')
parser.add_argument('--lr', type=float, default=None, metavar='LR',
help='learning rate (absolute lr)')
parser.add_argument('--blr', type=float, default=1e-3, metavar='LR',
help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
parser.add_argument('--layer_decay', type=float, default=0.75,
help='layer-wise lr decay from ELECTRA/BEiT')
parser.add_argument('--min_lr', type=float, default=1e-6, metavar='LR',
help='lower lr bound for cyclic schedulers that hit 0')
parser.add_argument('--warmup_epochs', type=int, default=5, metavar='N',
help='epochs to warmup LR')
# Augmentation parameters
parser.add_argument('--color_jitter', type=float, default=None, metavar='PCT',
help='Color jitter factor (enabled only when not using Auto/RandAug)')
parser.add_argument('--aa', type=str, default='rand-m9-mstd0.5-inc1', metavar='NAME',
help='Use AutoAugment policy. "v0" or "original". " + "(default: rand-m9-mstd0.5-inc1)'),
parser.add_argument('--smoothing', type=float, default=0.1,
help='Label smoothing (default: 0.1)')
# * Random Erase params
parser.add_argument('--reprob', type=float, default=0.25, metavar='PCT',
help='Random erase prob (default: 0.25)')
parser.add_argument('--remode', type=str, default='pixel',
help='Random erase mode (default: "pixel")')
parser.add_argument('--recount', type=int, default=1,
help='Random erase count (default: 1)')
parser.add_argument('--resplit', action='store_true', default=False,
help='Do not random erase first (clean) augmentation split')
# * Mixup params
parser.add_argument('--mixup', type=float, default=0,
help='mixup alpha, mixup enabled if > 0.')
parser.add_argument('--cutmix', type=float, default=0,
help='cutmix alpha, cutmix enabled if > 0.')
parser.add_argument('--cutmix_minmax', type=float, nargs='+', default=None,
help='cutmix min/max ratio, overrides alpha and enables cutmix if set (default: None)')
parser.add_argument('--mixup_prob', type=float, default=1.0,
help='Probability of performing mixup or cutmix when either/both is enabled')
parser.add_argument('--mixup_switch_prob', type=float, default=0.5,
help='Probability of switching to cutmix when both mixup and cutmix enabled')
parser.add_argument('--mixup_mode', type=str, default='batch',
help='How to apply mixup/cutmix params. Per "batch", "pair", or "elem"')
# * Finetuning params
parser.add_argument('--finetune', default='',
help='finetune from checkpoint')
parser.add_argument('--global_pool', action='store_true')
parser.set_defaults(global_pool=True)
parser.add_argument('--cls_token', action='store_false', dest='global_pool',
help='Use class token instead of global pool for classification')
# Dataset parameters
parser.add_argument('--data_path', default='/datasets01/imagenet_full_size/061417/', type=str,
help='dataset path')
parser.add_argument('--nb_classes', default=1000, type=int,
help='number of the classification types')
parser.add_argument('--output_dir', default='./output_dir',
help='path where to save, empty for no saving')
parser.add_argument('--log_dir', default='./output_dir',
help='path where to tensorboard log')
parser.add_argument('--device', default='cuda',
help='device to use for training / testing')
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--resume', default='',
help='resume from checkpoint')
parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
help='start epoch')
parser.add_argument('--eval', action='store_true',
help='Perform evaluation only')
parser.add_argument('--dist_eval', action='store_true', default=False,
help='Enabling distributed evaluation (recommended during training for faster monitor')
parser.add_argument('--num_workers', default=10, type=int)
parser.add_argument('--pin_mem', action='store_true',
help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
parser.add_argument('--no_pin_mem', action='store_false', dest='pin_mem')
parser.set_defaults(pin_mem=True)
# distributed training parameters
parser.add_argument('--world_size', default=1, type=int,
help='number of distributed processes')
parser.add_argument('--local_rank', default=-1, type=int)
parser.add_argument('--dist_on_itp', action='store_true')
parser.add_argument('--dist_url', default='env://',
help='url used to set up distributed training')
return parser
def main(args):
misc.init_distributed_mode(args)
print('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__))))
print("{}".format(args).replace(', ', ',\n'))
device = torch.device(args.device)
# fix the seed for reproducibility
seed = args.seed + misc.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
cudnn.benchmark = True
dataset_train = build_dataset(is_train=True, args=args)
dataset_val = build_dataset(is_train=False, args=args)
if True: # args.distributed:
num_tasks = misc.get_world_size()
global_rank = misc.get_rank()
sampler_train = torch.utils.data.DistributedSampler(
dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
)
print("Sampler_train = %s" % str(sampler_train))
if args.dist_eval:
if len(dataset_val) % num_tasks != 0:
print('Warning: Enabling distributed evaluation with an eval dataset not divisible by process number. '
'This will slightly alter validation results as extra duplicate entries are added to achieve '
'equal num of samples per-process.')
sampler_val = torch.utils.data.DistributedSampler(
dataset_val, num_replicas=num_tasks, rank=global_rank, shuffle=True) # shuffle=True to reduce monitor bias
else:
sampler_val = torch.utils.data.SequentialSampler(dataset_val)
else:
sampler_train = torch.utils.data.RandomSampler(dataset_train)
sampler_val = torch.utils.data.SequentialSampler(dataset_val)
if global_rank == 0 and args.log_dir is not None and not args.eval:
os.makedirs(args.log_dir, exist_ok=True)
log_writer = SummaryWriter(log_dir=args.log_dir)
else:
log_writer = None
data_loader_train = torch.utils.data.DataLoader(
dataset_train, sampler=sampler_train,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=True,
)
data_loader_val = torch.utils.data.DataLoader(
dataset_val, sampler=sampler_val,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=False
)
mixup_fn = None
mixup_active = args.mixup > 0 or args.cutmix > 0. or args.cutmix_minmax is not None
if mixup_active:
print("Mixup is activated!")
mixup_fn = Mixup(
mixup_alpha=args.mixup, cutmix_alpha=args.cutmix, cutmix_minmax=args.cutmix_minmax,
prob=args.mixup_prob, switch_prob=args.mixup_switch_prob, mode=args.mixup_mode,
label_smoothing=args.smoothing, num_classes=args.nb_classes)
model = models_vit.__dict__[args.model](
num_classes=args.nb_classes,
drop_path_rate=args.drop_path,
global_pool=args.global_pool,
)
if args.finetune and not args.eval:
checkpoint = torch.load(args.finetune, map_location='cpu')
print("Load pre-trained checkpoint from: %s" % args.finetune)
checkpoint_model = checkpoint['model']
state_dict = model.state_dict()
for k in ['head.weight', 'head.bias']:
if k in checkpoint_model and checkpoint_model[k].shape != state_dict[k].shape:
print(f"Removing key {k} from pretrained checkpoint")
del checkpoint_model[k]
# interpolate position embedding
interpolate_pos_embed(model, checkpoint_model)
# load pre-trained model
msg = model.load_state_dict(checkpoint_model, strict=False)
print(msg)
if args.global_pool:
assert set(msg.missing_keys) == {'head.weight', 'head.bias', 'fc_norm.weight', 'fc_norm.bias'}
else:
assert set(msg.missing_keys) == {'head.weight', 'head.bias'}
# manually initialize fc layer
trunc_normal_(model.head.weight, std=2e-5)
model.to(device)
model_without_ddp = model
n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
print("Model = %s" % str(model_without_ddp))
print('number of params (M): %.2f' % (n_parameters / 1.e6))
eff_batch_size = args.batch_size * args.accum_iter * misc.get_world_size()
if args.lr is None: # only base_lr is specified
args.lr = args.blr * eff_batch_size / 256
print("base lr: %.2e" % (args.lr * 256 / eff_batch_size))
print("actual lr: %.2e" % args.lr)
print("accumulate grad iterations: %d" % args.accum_iter)
print("effective batch size: %d" % eff_batch_size)
if args.distributed:
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
model_without_ddp = model.module
# build optimizer with layer-wise lr decay (lrd)
param_groups = lrd.param_groups_lrd(model_without_ddp, args.weight_decay,
no_weight_decay_list=model_without_ddp.no_weight_decay(),
layer_decay=args.layer_decay
)
optimizer = torch.optim.AdamW(param_groups, lr=args.lr)
loss_scaler = NativeScaler()
if mixup_fn is not None:
# smoothing is handled with mixup label transform
criterion = SoftTargetCrossEntropy()
elif args.smoothing > 0.:
criterion = LabelSmoothingCrossEntropy(smoothing=args.smoothing)
else:
criterion = torch.nn.CrossEntropyLoss()
print("criterion = %s" % str(criterion))
misc.load_model(args=args, model_without_ddp=model_without_ddp, optimizer=optimizer, loss_scaler=loss_scaler)
if args.eval:
test_stats = evaluate(data_loader_val, model, device)
print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']:.1f}%")
exit(0)
print(f"Start training for {args.epochs} epochs")
start_time = time.time()
max_accuracy = 0.0
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
data_loader_train.sampler.set_epoch(epoch)
train_stats = train_one_epoch(
model, criterion, data_loader_train,
optimizer, device, epoch, loss_scaler,
args.clip_grad, mixup_fn,
log_writer=log_writer,
args=args
)
if args.output_dir:
misc.save_model(
args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
loss_scaler=loss_scaler, epoch=epoch)
test_stats = evaluate(data_loader_val, model, device)
print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']:.1f}%")
max_accuracy = max(max_accuracy, test_stats["acc1"])
print(f'Max accuracy: {max_accuracy:.2f}%')
if log_writer is not None:
log_writer.add_scalar('perf/test_acc1', test_stats['acc1'], epoch)
log_writer.add_scalar('perf/test_acc5', test_stats['acc5'], epoch)
log_writer.add_scalar('perf/test_loss', test_stats['loss'], epoch)
log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
**{f'test_{k}': v for k, v in test_stats.items()},
'epoch': epoch,
'n_parameters': n_parameters}
if args.output_dir and misc.is_main_process():
if log_writer is not None:
log_writer.flush()
with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
f.write(json.dumps(log_stats) + "\n")
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
print('Training time {}'.format(total_time_str))
if __name__ == '__main__':
args = get_args_parser()
args = args.parse_args()
if args.output_dir:
Path(args.output_dir).mkdir(parents=True, exist_ok=True)
main(args)
PyTorch/NLP/Vision_Transformer/main_linprobe.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# MoCo v3: https://github.com/facebookresearch/moco-v3
# --------------------------------------------------------
import argparse
import datetime
import json
import numpy as np
import os
import time
from pathlib import Path
import torch
import torch.backends.cudnn as cudnn
from torch.utils.tensorboard import SummaryWriter
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import timm
assert timm.__version__ == "0.3.2" # version check
from timm.models.layers import trunc_normal_
import util.misc as misc
from util.pos_embed import interpolate_pos_embed
from util.misc import NativeScalerWithGradNormCount as NativeScaler
from util.lars import LARS
from util.crop import RandomResizedCrop
import models_vit
from engine_finetune import train_one_epoch, evaluate
def get_args_parser():
parser = argparse.ArgumentParser('MAE linear probing for image classification', add_help=False)
parser.add_argument('--batch_size', default=512, type=int,
help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
parser.add_argument('--epochs', default=90, type=int)
parser.add_argument('--accum_iter', default=1, type=int,
help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')
# Model parameters
parser.add_argument('--model', default='vit_large_patch16', type=str, metavar='MODEL',
help='Name of model to train')
# Optimizer parameters
parser.add_argument('--weight_decay', type=float, default=0,
help='weight decay (default: 0 for linear probe following MoCo v1)')
parser.add_argument('--lr', type=float, default=None, metavar='LR',
help='learning rate (absolute lr)')
parser.add_argument('--blr', type=float, default=0.1, metavar='LR',
help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
parser.add_argument('--min_lr', type=float, default=0., metavar='LR',
help='lower lr bound for cyclic schedulers that hit 0')
parser.add_argument('--warmup_epochs', type=int, default=10, metavar='N',
help='epochs to warmup LR')
# * Finetuning params
parser.add_argument('--finetune', default='',
help='finetune from checkpoint')
parser.add_argument('--global_pool', action='store_true')
parser.set_defaults(global_pool=False)
parser.add_argument('--cls_token', action='store_false', dest='global_pool',
help='Use class token instead of global pool for classification')
# Dataset parameters
parser.add_argument('--data_path', default='/datasets01/imagenet_full_size/061417/', type=str,
help='dataset path')
parser.add_argument('--nb_classes', default=1000, type=int,
help='number of the classification types')
parser.add_argument('--output_dir', default='./output_dir',
help='path where to save, empty for no saving')
parser.add_argument('--log_dir', default='./output_dir',
help='path where to tensorboard log')
parser.add_argument('--device', default='cuda',
help='device to use for training / testing')
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--resume', default='',
help='resume from checkpoint')
parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
help='start epoch')
parser.add_argument('--eval', action='store_true',
help='Perform evaluation only')
parser.add_argument('--dist_eval', action='store_true', default=False,
help='Enabling distributed evaluation (recommended during training for faster monitor')
parser.add_argument('--num_workers', default=10, type=int)
parser.add_argument('--pin_mem', action='store_true',
help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
parser.add_argument('--no_pin_mem', action='store_false', dest='pin_mem')
parser.set_defaults(pin_mem=True)
# distributed training parameters
parser.add_argument('--world_size', default=1, type=int,
help='number of distributed processes')
parser.add_argument('--local_rank', default=-1, type=int)
parser.add_argument('--dist_on_itp', action='store_true')
parser.add_argument('--dist_url', default='env://',
help='url used to set up distributed training')
return parser
def main(args):
misc.init_distributed_mode(args)
print('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__))))
print("{}".format(args).replace(', ', ',\n'))
device = torch.device(args.device)
# fix the seed for reproducibility
seed = args.seed + misc.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
cudnn.benchmark = True
# linear probe: weak augmentation
transform_train = transforms.Compose([
RandomResizedCrop(224, interpolation=3),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
transform_val = transforms.Compose([
transforms.Resize(256, interpolation=3),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
dataset_train = datasets.ImageFolder(os.path.join(args.data_path, 'train'), transform=transform_train)
dataset_val = datasets.ImageFolder(os.path.join(args.data_path, 'val'), transform=transform_val)
print(dataset_train)
print(dataset_val)
if True: # args.distributed:
num_tasks = misc.get_world_size()
global_rank = misc.get_rank()
sampler_train = torch.utils.data.DistributedSampler(
dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
)
print("Sampler_train = %s" % str(sampler_train))
if args.dist_eval:
if len(dataset_val) % num_tasks != 0:
print('Warning: Enabling distributed evaluation with an eval dataset not divisible by process number. '
'This will slightly alter validation results as extra duplicate entries are added to achieve '
'equal num of samples per-process.')
sampler_val = torch.utils.data.DistributedSampler(
dataset_val, num_replicas=num_tasks, rank=global_rank, shuffle=True) # shuffle=True to reduce monitor bias
else:
sampler_val = torch.utils.data.SequentialSampler(dataset_val)
else:
sampler_train = torch.utils.data.RandomSampler(dataset_train)
sampler_val = torch.utils.data.SequentialSampler(dataset_val)
if global_rank == 0 and args.log_dir is not None and not args.eval:
os.makedirs(args.log_dir, exist_ok=True)
log_writer = SummaryWriter(log_dir=args.log_dir)
else:
log_writer = None
data_loader_train = torch.utils.data.DataLoader(
dataset_train, sampler=sampler_train,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=True,
)
data_loader_val = torch.utils.data.DataLoader(
dataset_val, sampler=sampler_val,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=False
)
model = models_vit.__dict__[args.model](
num_classes=args.nb_classes,
global_pool=args.global_pool,
)
if args.finetune and not args.eval:
checkpoint = torch.load(args.finetune, map_location='cpu')
print("Load pre-trained checkpoint from: %s" % args.finetune)
checkpoint_model = checkpoint['model']
state_dict = model.state_dict()
for k in ['head.weight', 'head.bias']:
if k in checkpoint_model and checkpoint_model[k].shape != state_dict[k].shape:
print(f"Removing key {k} from pretrained checkpoint")
del checkpoint_model[k]
# interpolate position embedding
interpolate_pos_embed(model, checkpoint_model)
# load pre-trained model
msg = model.load_state_dict(checkpoint_model, strict=False)
print(msg)
if args.global_pool:
assert set(msg.missing_keys) == {'head.weight', 'head.bias', 'fc_norm.weight', 'fc_norm.bias'}
else:
assert set(msg.missing_keys) == {'head.weight', 'head.bias'}
# manually initialize fc layer: following MoCo v3
trunc_normal_(model.head.weight, std=0.01)
# for linear prob only
# hack: revise model's head with BN
model.head = torch.nn.Sequential(torch.nn.BatchNorm1d(model.head.in_features, affine=False, eps=1e-6), model.head)
# freeze all but the head
for _, p in model.named_parameters():
p.requires_grad = False
for _, p in model.head.named_parameters():
p.requires_grad = True
model.to(device)
model_without_ddp = model
n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
print("Model = %s" % str(model_without_ddp))
print('number of params (M): %.2f' % (n_parameters / 1.e6))
eff_batch_size = args.batch_size * args.accum_iter * misc.get_world_size()
if args.lr is None: # only base_lr is specified
args.lr = args.blr * eff_batch_size / 256
print("base lr: %.2e" % (args.lr * 256 / eff_batch_size))
print("actual lr: %.2e" % args.lr)
print("accumulate grad iterations: %d" % args.accum_iter)
print("effective batch size: %d" % eff_batch_size)
if args.distributed:
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
model_without_ddp = model.module
optimizer = LARS(model_without_ddp.head.parameters(), lr=args.lr, weight_decay=args.weight_decay)
print(optimizer)
loss_scaler = NativeScaler()
criterion = torch.nn.CrossEntropyLoss()
print("criterion = %s" % str(criterion))
misc.load_model(args=args, model_without_ddp=model_without_ddp, optimizer=optimizer, loss_scaler=loss_scaler)
if args.eval:
test_stats = evaluate(data_loader_val, model, device)
print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']:.1f}%")
exit(0)
print(f"Start training for {args.epochs} epochs")
start_time = time.time()
max_accuracy = 0.0
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
data_loader_train.sampler.set_epoch(epoch)
train_stats = train_one_epoch(
model, criterion, data_loader_train,
optimizer, device, epoch, loss_scaler,
max_norm=None,
log_writer=log_writer,
args=args
)
if args.output_dir:
misc.save_model(
args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
loss_scaler=loss_scaler, epoch=epoch)
test_stats = evaluate(data_loader_val, model, device)
print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']:.1f}%")
max_accuracy = max(max_accuracy, test_stats["acc1"])
print(f'Max accuracy: {max_accuracy:.2f}%')
if log_writer is not None:
log_writer.add_scalar('perf/test_acc1', test_stats['acc1'], epoch)
log_writer.add_scalar('perf/test_acc5', test_stats['acc5'], epoch)
log_writer.add_scalar('perf/test_loss', test_stats['loss'], epoch)
log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
**{f'test_{k}': v for k, v in test_stats.items()},
'epoch': epoch,
'n_parameters': n_parameters}
if args.output_dir and misc.is_main_process():
if log_writer is not None:
log_writer.flush()
with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
f.write(json.dumps(log_stats) + "\n")
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
print('Training time {}'.format(total_time_str))
if __name__ == '__main__':
args = get_args_parser()
args = args.parse_args()
if args.output_dir:
Path(args.output_dir).mkdir(parents=True, exist_ok=True)
main(args)
PyTorch/NLP/Vision_Transformer/main_pretrain.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# DeiT: https://github.com/facebookresearch/deit
# BEiT: https://github.com/microsoft/unilm/tree/master/beit
# --------------------------------------------------------
import argparse
import datetime
import json
import numpy as np
import os
import time
from pathlib import Path
import torch
import torch.backends.cudnn as cudnn
from torch.utils.tensorboard import SummaryWriter
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import timm
assert timm.__version__ == "0.3.2" # version check
import timm.optim.optim_factory as optim_factory
import util.misc as misc
from util.misc import NativeScalerWithGradNormCount as NativeScaler
import models_mae
from engine_pretrain import train_one_epoch
def get_args_parser():
parser = argparse.ArgumentParser('MAE pre-training', add_help=False)
parser.add_argument('--batch_size', default=64, type=int,
help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
parser.add_argument('--epochs', default=400, type=int)
parser.add_argument('--accum_iter', default=1, type=int,
help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')
# Model parameters
parser.add_argument('--model', default='mae_vit_large_patch16', type=str, metavar='MODEL',
help='Name of model to train')
parser.add_argument('--input_size', default=224, type=int,
help='images input size')
parser.add_argument('--mask_ratio', default=0.75, type=float,
help='Masking ratio (percentage of removed patches).')
parser.add_argument('--norm_pix_loss', action='store_true',
help='Use (per-patch) normalized pixels as targets for computing loss')
parser.set_defaults(norm_pix_loss=False)
# Optimizer parameters
parser.add_argument('--weight_decay', type=float, default=0.05,
help='weight decay (default: 0.05)')
parser.add_argument('--lr', type=float, default=None, metavar='LR',
help='learning rate (absolute lr)')
parser.add_argument('--blr', type=float, default=1e-3, metavar='LR',
help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
parser.add_argument('--min_lr', type=float, default=0., metavar='LR',
help='lower lr bound for cyclic schedulers that hit 0')
parser.add_argument('--warmup_epochs', type=int, default=40, metavar='N',
help='epochs to warmup LR')
# Dataset parameters
parser.add_argument('--data_path', default='/datasets01/imagenet_full_size/061417/', type=str,
help='dataset path')
parser.add_argument('--output_dir', default='./output_dir',
help='path where to save, empty for no saving')
parser.add_argument('--log_dir', default='./output_dir',
help='path where to tensorboard log')
parser.add_argument('--device', default='cuda',
help='device to use for training / testing')
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--resume', default='',
help='resume from checkpoint')
parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
help='start epoch')
parser.add_argument('--num_workers', default=10, type=int)
parser.add_argument('--pin_mem', action='store_true',
help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
parser.add_argument('--no_pin_mem', action='store_false', dest='pin_mem')
parser.set_defaults(pin_mem=True)
# distributed training parameters
parser.add_argument('--world_size', default=1, type=int,
help='number of distributed processes')
parser.add_argument('--local_rank', default=-1, type=int)
parser.add_argument('--dist_on_itp', action='store_true')
parser.add_argument('--dist_url', default='env://',
help='url used to set up distributed training')
return parser
def main(args):
misc.init_distributed_mode(args)
print('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__))))
print("{}".format(args).replace(', ', ',\n'))
device = torch.device(args.device)
# fix the seed for reproducibility
seed = args.seed + misc.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
cudnn.benchmark = True
# simple augmentation
transform_train = transforms.Compose([
transforms.RandomResizedCrop(args.input_size, scale=(0.2, 1.0), interpolation=3), # 3 is bicubic
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
dataset_train = datasets.ImageFolder(os.path.join(args.data_path, 'train'), transform=transform_train)
print(dataset_train)
if True: # args.distributed:
num_tasks = misc.get_world_size()
global_rank = misc.get_rank()
sampler_train = torch.utils.data.DistributedSampler(
dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
)
print("Sampler_train = %s" % str(sampler_train))
else:
sampler_train = torch.utils.data.RandomSampler(dataset_train)
if global_rank == 0 and args.log_dir is not None:
os.makedirs(args.log_dir, exist_ok=True)
log_writer = SummaryWriter(log_dir=args.log_dir)
else:
log_writer = None
data_loader_train = torch.utils.data.DataLoader(
dataset_train, sampler=sampler_train,
batch_size=args.batch_size,
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=True,
)
# define the model
model = models_mae.__dict__[args.model](norm_pix_loss=args.norm_pix_loss)
model.to(device)
model_without_ddp = model
print("Model = %s" % str(model_without_ddp))
eff_batch_size = args.batch_size * args.accum_iter * misc.get_world_size()
if args.lr is None: # only base_lr is specified
args.lr = args.blr * eff_batch_size / 256
print("base lr: %.2e" % (args.lr * 256 / eff_batch_size))
print("actual lr: %.2e" % args.lr)
print("accumulate grad iterations: %d" % args.accum_iter)
print("effective batch size: %d" % eff_batch_size)
if args.distributed:
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu], find_unused_parameters=True)
model_without_ddp = model.module
# following timm: set wd as 0 for bias and norm layers
param_groups = optim_factory.add_weight_decay(model_without_ddp, args.weight_decay)
optimizer = torch.optim.AdamW(param_groups, lr=args.lr, betas=(0.9, 0.95))
print(optimizer)
loss_scaler = NativeScaler()
misc.load_model(args=args, model_without_ddp=model_without_ddp, optimizer=optimizer, loss_scaler=loss_scaler)
print(f"Start training for {args.epochs} epochs")
start_time = time.time()
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
data_loader_train.sampler.set_epoch(epoch)
train_stats = train_one_epoch(
model, data_loader_train,
optimizer, device, epoch, loss_scaler,
log_writer=log_writer,
args=args
)
if args.output_dir and (epoch % 20 == 0 or epoch + 1 == args.epochs):
misc.save_model(
args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
loss_scaler=loss_scaler, epoch=epoch)
log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
'epoch': epoch,}
if args.output_dir and misc.is_main_process():
if log_writer is not None:
log_writer.flush()
with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
f.write(json.dumps(log_stats) + "\n")
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
print('Training time {}'.format(total_time_str))
if __name__ == '__main__':
args = get_args_parser()
args = args.parse_args()
if args.output_dir:
Path(args.output_dir).mkdir(parents=True, exist_ok=True)
main(args)
PyTorch/NLP/Vision_Transformer/models_mae.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# timm: https://github.com/rwightman/pytorch-image-models/tree/master/timm
# DeiT: https://github.com/facebookresearch/deit
# --------------------------------------------------------
from functools import partial
import torch
import torch.nn as nn
from timm.models.vision_transformer import PatchEmbed, Block
from util.pos_embed import get_2d_sincos_pos_embed
class MaskedAutoencoderViT(nn.Module):
""" Masked Autoencoder with VisionTransformer backbone
"""
def __init__(self, img_size=224, patch_size=16, in_chans=3,
embed_dim=1024, depth=24, num_heads=16,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4., norm_layer=nn.LayerNorm, norm_pix_loss=False):
super().__init__()
# --------------------------------------------------------------------------
# MAE encoder specifics
self.patch_embed = PatchEmbed(img_size, patch_size, in_chans, embed_dim)
num_patches = self.patch_embed.num_patches
self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim), requires_grad=False) # fixed sin-cos embedding
self.blocks = nn.ModuleList([
Block(embed_dim, num_heads, mlp_ratio, qkv_bias=True, qk_scale=None, norm_layer=norm_layer)
for i in range(depth)])
self.norm = norm_layer(embed_dim)
# --------------------------------------------------------------------------
# --------------------------------------------------------------------------
# MAE decoder specifics
self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
self.mask_token = nn.Parameter(torch.zeros(1, 1, decoder_embed_dim))
self.decoder_pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, decoder_embed_dim), requires_grad=False) # fixed sin-cos embedding
self.decoder_blocks = nn.ModuleList([
Block(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, qk_scale=None, norm_layer=norm_layer)
for i in range(decoder_depth)])
self.decoder_norm = norm_layer(decoder_embed_dim)
self.decoder_pred = nn.Linear(decoder_embed_dim, patch_size**2 * in_chans, bias=True) # decoder to patch
# --------------------------------------------------------------------------
self.norm_pix_loss = norm_pix_loss
self.initialize_weights()
def initialize_weights(self):
# initialization
# initialize (and freeze) pos_embed by sin-cos embedding
pos_embed = get_2d_sincos_pos_embed(self.pos_embed.shape[-1], int(self.patch_embed.num_patches**.5), cls_token=True)
self.pos_embed.data.copy_(torch.from_numpy(pos_embed).float().unsqueeze(0))
decoder_pos_embed = get_2d_sincos_pos_embed(self.decoder_pos_embed.shape[-1], int(self.patch_embed.num_patches**.5), cls_token=True)
self.decoder_pos_embed.data.copy_(torch.from_numpy(decoder_pos_embed).float().unsqueeze(0))
# initialize patch_embed like nn.Linear (instead of nn.Conv2d)
w = self.patch_embed.proj.weight.data
torch.nn.init.xavier_uniform_(w.view([w.shape[0], -1]))
# timm's trunc_normal_(std=.02) is effectively normal_(std=0.02) as cutoff is too big (2.)
torch.nn.init.normal_(self.cls_token, std=.02)
torch.nn.init.normal_(self.mask_token, std=.02)
# initialize nn.Linear and nn.LayerNorm
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
# we use xavier_uniform following official JAX ViT:
torch.nn.init.xavier_uniform_(m.weight)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
def patchify(self, imgs):
"""
imgs: (N, 3, H, W)
x: (N, L, patch_size**2 *3)
"""
p = self.patch_embed.patch_size[0]
assert imgs.shape[2] == imgs.shape[3] and imgs.shape[2] % p == 0
h = w = imgs.shape[2] // p
x = imgs.reshape(shape=(imgs.shape[0], 3, h, p, w, p))
x = torch.einsum('nchpwq->nhwpqc', x)
x = x.reshape(shape=(imgs.shape[0], h * w, p**2 * 3))
return x
def unpatchify(self, x):
"""
x: (N, L, patch_size**2 *3)
imgs: (N, 3, H, W)
"""
p = self.patch_embed.patch_size[0]
h = w = int(x.shape[1]**.5)
assert h * w == x.shape[1]
x = x.reshape(shape=(x.shape[0], h, w, p, p, 3))
x = torch.einsum('nhwpqc->nchpwq', x)
imgs = x.reshape(shape=(x.shape[0], 3, h * p, h * p))
return imgs
def random_masking(self, x, mask_ratio):
"""
Perform per-sample random masking by per-sample shuffling.
Per-sample shuffling is done by argsort random noise.
x: [N, L, D], sequence
"""
N, L, D = x.shape # batch, length, dim
len_keep = int(L * (1 - mask_ratio))
noise = torch.rand(N, L, device=x.device) # noise in [0, 1]
# sort noise for each sample
ids_shuffle = torch.argsort(noise, dim=1) # ascend: small is keep, large is remove
ids_restore = torch.argsort(ids_shuffle, dim=1)
# keep the first subset
ids_keep = ids_shuffle[:, :len_keep]
x_masked = torch.gather(x, dim=1, index=ids_keep.unsqueeze(-1).repeat(1, 1, D))
# generate the binary mask: 0 is keep, 1 is remove
mask = torch.ones([N, L], device=x.device)
mask[:, :len_keep] = 0
# unshuffle to get the binary mask
mask = torch.gather(mask, dim=1, index=ids_restore)
return x_masked, mask, ids_restore
def forward_encoder(self, x, mask_ratio):
# embed patches
x = self.patch_embed(x)
# add pos embed w/o cls token
x = x + self.pos_embed[:, 1:, :]
# masking: length -> length * mask_ratio
x, mask, ids_restore = self.random_masking(x, mask_ratio)
# append cls token
cls_token = self.cls_token + self.pos_embed[:, :1, :]
cls_tokens = cls_token.expand(x.shape[0], -1, -1)
x = torch.cat((cls_tokens, x), dim=1)
# apply Transformer blocks
for blk in self.blocks:
x = blk(x)
x = self.norm(x)
return x, mask, ids_restore
def forward_decoder(self, x, ids_restore):
# embed tokens
x = self.decoder_embed(x)
# append mask tokens to sequence
mask_tokens = self.mask_token.repeat(x.shape[0], ids_restore.shape[1] + 1 - x.shape[1], 1)
x_ = torch.cat([x[:, 1:, :], mask_tokens], dim=1) # no cls token
x_ = torch.gather(x_, dim=1, index=ids_restore.unsqueeze(-1).repeat(1, 1, x.shape[2])) # unshuffle
x = torch.cat([x[:, :1, :], x_], dim=1) # append cls token
# add pos embed
x = x + self.decoder_pos_embed
# apply Transformer blocks
for blk in self.decoder_blocks:
x = blk(x)
x = self.decoder_norm(x)
# predictor projection
x = self.decoder_pred(x)
# remove cls token
x = x[:, 1:, :]
return x
def forward_loss(self, imgs, pred, mask):
"""
imgs: [N, 3, H, W]
pred: [N, L, p*p*3]
mask: [N, L], 0 is keep, 1 is remove,
"""
target = self.patchify(imgs)
if self.norm_pix_loss:
mean = target.mean(dim=-1, keepdim=True)
var = target.var(dim=-1, keepdim=True)
target = (target - mean) / (var + 1.e-6)**.5
loss = (pred - target) ** 2
loss = loss.mean(dim=-1) # [N, L], mean loss per patch
loss = (loss * mask).sum() / mask.sum() # mean loss on removed patches
return loss
def forward(self, imgs, mask_ratio=0.75):
latent, mask, ids_restore = self.forward_encoder(imgs, mask_ratio)
pred = self.forward_decoder(latent, ids_restore) # [N, L, p*p*3]
loss = self.forward_loss(imgs, pred, mask)
return loss, pred, mask
def mae_vit_base_patch16_dec512d8b(**kwargs):
model = MaskedAutoencoderViT(
patch_size=16, embed_dim=768, depth=12, num_heads=12,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def mae_vit_large_patch16_dec512d8b(**kwargs):
model = MaskedAutoencoderViT(
patch_size=16, embed_dim=1024, depth=24, num_heads=16,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def mae_vit_huge_patch14_dec512d8b(**kwargs):
model = MaskedAutoencoderViT(
patch_size=14, embed_dim=1280, depth=32, num_heads=16,
decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
# set recommended archs
mae_vit_base_patch16 = mae_vit_base_patch16_dec512d8b # decoder: 512 dim, 8 blocks
mae_vit_large_patch16 = mae_vit_large_patch16_dec512d8b # decoder: 512 dim, 8 blocks
mae_vit_huge_patch14 = mae_vit_huge_patch14_dec512d8b # decoder: 512 dim, 8 blocks
PyTorch/NLP/Vision_Transformer/models_vit.py 0 → 100644
View file @ c056df78
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# timm: https://github.com/rwightman/pytorch-image-models/tree/master/timm
# DeiT: https://github.com/facebookresearch/deit
# --------------------------------------------------------
from functools import partial
import torch
import torch.nn as nn
import timm.models.vision_transformer
class VisionTransformer(timm.models.vision_transformer.VisionTransformer):
""" Vision Transformer with support for global average pooling
"""
def __init__(self, global_pool=False, **kwargs):
super(VisionTransformer, self).__init__(**kwargs)
self.global_pool = global_pool
if self.global_pool:
norm_layer = kwargs['norm_layer']
embed_dim = kwargs['embed_dim']
self.fc_norm = norm_layer(embed_dim)
del self.norm # remove the original norm
def forward_features(self, x):
B = x.shape[0]
x = self.patch_embed(x)
cls_tokens = self.cls_token.expand(B, -1, -1) # stole cls_tokens impl from Phil Wang, thanks
x = torch.cat((cls_tokens, x), dim=1)
x = x + self.pos_embed
x = self.pos_drop(x)
for blk in self.blocks:
x = blk(x)
if self.global_pool:
x = x[:, 1:, :].mean(dim=1) # global pool without cls token
outcome = self.fc_norm(x)
else:
x = self.norm(x)
outcome = x[:, 0]
return outcome
def vit_base_patch16(**kwargs):
model = VisionTransformer(
patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def vit_large_patch16(**kwargs):
model = VisionTransformer(
patch_size=16, embed_dim=1024, depth=24, num_heads=16, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
def vit_huge_patch14(**kwargs):
model = VisionTransformer(
patch_size=14, embed_dim=1280, depth=32, num_heads=16, mlp_ratio=4, qkv_bias=True,
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
return model
\ No newline at end of file
PyTorch/NLP/Vision_Transformer/run-eval.sh 0 → 100644
View file @ c056df78
export PRETRAIN_CHKPT=/work/home/hepj/model/VIT/mae_finetuned_vit_base.pth
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
python3 main_finetune.py\
--eval\
--resume ${PRETRAIN_CHKPT}\
--model vit_base_patch16\
--batch_size 16\
--data_path ${IMAGENET_DIR}
PyTorch/NLP/Vision_Transformer/run-finetune-4.sh 0 → 100644
View file @ c056df78
export HIP_VISIBLE_DEVICE=0,1,2,3
export PRETRAIN_CHKPT=/work/home/hepj/model/VIT/mae_pretrain_vit_base.pth
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
OMP_NUM_THREADS=1 python3 -m torch.distributed.launch --nproc_per_node=4 main_finetune.py \
--accum_iter 4 \
--batch_size 32 \
--model vit_base_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 1 \
--blr 5e-4 --layer_decay 0.65 \
--weight_decay 0.05 --drop_path 0.1 --mixup 0.8 --cutmix 1.0 --reprob 0.25 \
--dist_eval --data_path ${IMAGENET_DIR}
PyTorch/NLP/Vision_Transformer/run-finetune.sh 0 → 100644
View file @ c056df78
export HIP_VISIBLE_DEVICE=0
export PRETRAIN_CHKPT=/work/home/hepj/model/VIT/mae_pretrain_vit_base.pth
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
python3 main_finetune.py \
--batch_size 32 \
--model vit_base_patch16 \
--finetune ${PRETRAIN_CHKPT} \
--epochs 1 \
--blr 5e-4 --layer_decay 0.65 \
--weight_decay 0.05 --drop_path 0.1 --mixup 0.8 --cutmix 1.0 --reprob 0.25 \
--dist_eval --data_path ${IMAGENET_DIR}
PyTorch/NLP/Vision_Transformer/run-pretrain-4.sh 0 → 100644
View file @ c056df78
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
export HIP_VISIBLE_DEVICE=0,1,2,3
OMP_NUM_THREADS=1 python3 -m torch.distributed.launch --nproc_per_node=4 main_pretrain.py --epochs 1 --accum_iter 4 --model mae_vit_base_patch16 --batch_size 64 --model mae_vit_base_patch16 --data_path ${IMAGENET_DIR}
PyTorch/NLP/Vision_Transformer/run-pretrain.sh 0 → 100644
View file @ c056df78
#!/bin/bash
export HIP_VISIBLE_DEVICE=0
export IMAGENET_DIR=/public/DL_DATA/ImageNet-pytorch
python3 main_pretrain.py --epochs 1 --model mae_vit_base_patch16 --batch_size 64 --model mae_vit_base_patch16 --data_path ${IMAGENET_DIR}
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