Skip to content

Commit 9fe85ec

Browse files
lmk123568lmk123568
authored
ref readme (#102)
1 parent 8ab66db commit 9fe85ec

File tree

4 files changed

+218
-21
lines changed

4 files changed

+218
-21
lines changed

‎README.md

+99-21
Original file line numberDiff line numberDiff line change
@@ -1,31 +1,109 @@
1-
# PASSL
1+
⚙️ English | [简体中文](./README_cn.md)
2+
3+
<p align="center">
4+
<img src="./docs/imgs/passl_logo.svg" width="60%" height="60%"/>
5+
</p>
6+
<p align="center">
7+
<a href="./LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-red.svg"></a>
8+
<a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
9+
<a href="https://github.com/PaddlePaddle/PASSL/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PASSL?color=ccf"></a>
10+
<a href=""><img src="https://camo.githubusercontent.com/abb97269de2982c379cbc128bba93ba724d8822bfbe082737772bd4feb59cb54/68747470733a2f2f63646e2e7261776769742e636f6d2f73696e647265736f726875732f617765736f6d652f643733303566333864323966656437386661383536353265336136336531353464643865383832392f6d656469612f62616467652e737667"></a>
11+
<a href="https://aistudio.baidu.com/aistudio/personalcenter/thirdview/940489"><img src="https://img.shields.io/badge/Tutorial-AI Studio-blue.svg"></a>
12+
</p>
213

314
## Introduction
15+
416
PASSL is a Paddle based vision library for state-of-the-art Self-Supervised Learning research with [PaddlePaddle](https://www.paddlepaddle.org.cn/). PASSL aims to accelerate research cycle in self-supervised learning: from designing a new self-supervised task to evaluating the learned representations.
5-
- **Reproducible implementation of SOTA in Self-Supervision**: Existing SOTA in Self-Supervision are implemented - [SimCLR](https://arxiv.org/abs/2002.05709), [MoCo(v1)](https://arxiv.org/abs/1911.05722),[MoCo(v2)](https://arxiv.org/abs/1911.05722), [MoCo-BYOL](docs/Train_MoCo-BYOL_model.md), [CLIP](https://arxiv.org/abs/2103.00020). [BYOL](https://arxiv.org/abs/2006.07733) is coming soon. Also supports supervised trainings.
6-
- **Modular**: Easy to build new tasks and reuse the existing components from other tasks (Trainer, models and heads, data transforms, etc.).
717

8-
## Installation
9-
- See [INSTALL.md](https://github.com/PaddlePaddle/PASSL/blob/main/docs/INSTALL.md).
18+
Key features of PASSL：
19+
20+
- Reproducible implementation of SOTA in Self-Supervision
21+
22+
Existing SOTA in Self-Supervision are implemented - [SimCLR](https://arxiv.org/abs/2002.05709), [MoCo(v1)](https://arxiv.org/abs/1911.05722), [MoCo(v2)](https://arxiv.org/abs/1911.05722), [MoCo-BYOL](docs/Train_MoCo-BYOL_model.md), [BYOL](https://arxiv.org/abs/2006.07733), [BEiT](https://arxiv.org/abs/2106.08254). Supervised classification training is also supported.
23+
24+
- Modular Design
25+
26+
Easy to build new tasks and reuse the existing components from other tasks (Trainer, models and heads, data transforms, etc.)
27+
28+
🛠️ The ultimate goal of PASSL is to use self-supervised learning to provide more appropriate pre-training weights for downstream tasks while significantly reducing the cost of data annotation.
29+
30+
**📣 Recent Update:**
31+
32+
* (2022-2-9): Refactoring README
33+
* 🔥 Now：
1034

1135
## Implemented Models
12-
Benchmark Linear Image Classification on ImageNet-1K
13-
| | epochs |official results | passl results | Backbone| Model |
14-
| ---|--- | ---- | ---- | ----| ---- |
15-
| MoCo | 200 | 60.6| 60.64| ResNet-50 | [download](https://passl.bj.bcebos.com/models/moco_v1_r50_e200_ckpt.pdparams)|
16-
| SimCLR | 100 | 64.5 | 65.3 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/simclr_r50_ep100_ckpt.pdparams)|
17-
| MoCo v2 | 200 | 67.7 | 67.72| ResNet-50 | [download](https://passl.bj.bcebos.com/models/moco_v2_r50_e200_ckpt.pdparams)|
18-
| MoCo-BYOL | 300 | 71.56 | 72.10| ResNet-50 | [download](https://passl.bj.bcebos.com/models/mocobyol_r50_ep300_ckpt.pdparams)|
19-
| BYOL | 300 | 72.50 | 71.62| ResNet-50 | [download](https://passl.bj.bcebos.com/models/byol_r50_300.pdparams)|
20-
| PixPro | 100 | 55.1(fp16) | 57.2(fp32) | ResNet-50 | [download](https://passl.bj.bcebos.com/models/pixpro_r50_ep100_no_instance_with_linear.pdparams)
36+
37+
* **Self-Supervised Learning Models**
38+
39+
PASSL implements a series of self-supervised learning algorithms, See **Document** for details on its use
40+
41+
| | Epochs | Official results | PASSL results | Backbone | Model | Document |
42+
| --------- | ------ | ---------------- | ------------- | --------- | ------------------------------------------------------------ | ------------------------------------------------ |
43+
| MoCo | 200 | 60.6 | 60.64 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/moco_v1_r50_e200_ckpt.pdparams) | [Train MoCo](docs/Train_MoCo_model.md) |
44+
| SimCLR | 100 | 64.5 | 65.3 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/simclr_r50_ep100_ckpt.pdparams) | [Train SimCLR](docs/Train_SimCLR_model.md) |
45+
| MoCo v2 | 200 | 67.7 | 67.72 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/moco_v2_r50_e200_ckpt.pdparams) | [Train MoCo](docs/Train_MoCo_model.md) |
46+
| MoCo-BYOL | 300 | 71.56 | 72.10 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/mocobyol_r50_ep300_ckpt.pdparams) | [Train MoCo-BYOL](docs/Train_MoCo-BYOL_model.md) |
47+
| BYOL | 300 | 72.50 | 71.62 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/byol_r50_300.pdparams) | [Train BYOL](docs/Train_BYOL_model.md) |
48+
| PixPro | 100 | 55.1(fp16) | 57.2(fp32) | ResNet-50 | [download](https://passl.bj.bcebos.com/models/pixpro_r50_ep100_no_instance_with_linear.pdparams) | [Train PixPro](docs/Train_PixPro_model.md) |
49+
50+
> Benchmark Linear Image Classification on ImageNet-1K.
51+
52+
Comming Soon：More algorithm implementations are already in our plans ...
53+
54+
* **Classification Models**
55+
56+
PASSL implements influential image classification algorithms such as Visual Transformer, and provides corresponding pre-training weights. Designed to support the construction and research of self-supervised, multimodal, large-model algorithms. See [Classification_Models_Guide.md](docs/Classification_Models_Guide.md) for more usage details
57+
58+
| | Detail | Tutorial |
59+
| ---------------- | --------------------------- | ------------------------------------------------------------ |
60+
| ViT | / | [PaddleEdu](https://aistudio.baidu.com/aistudio/projectdetail/2293050) |
61+
| Swin Transformer | / | [PaddleEdu](https://aistudio.baidu.com/aistudio/projectdetail/2280436) |
62+
| CaiT | [config](configs/cait) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401469) |
63+
| T2T-ViT | [config](configs/t2t_vit) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401348) |
64+
| CvT | [config](configs/cvt) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401386) |
65+
| BEiT | [config](configs/beit) | [unofficial](https://aistudio.baidu.com/aistudio/projectdetail/2417241) |
66+
| MLP-Mixer | [config](configs/mlp_mixer) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401295) |
67+
| ConvNeXt | [config](configs/convnext) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3407445) |
68+
69+
🔥 PASSL provides a detailed dissection of the algorithm, see **Tutorial** for details.
70+
71+
## Installation
72+
73+
See [INSTALL.md](https://github.com/PaddlePaddle/PASSL/blob/main/docs/INSTALL.md).
2174

2275
## Getting Started
76+
2377
Please see [GETTING_STARTED.md](https://github.com/PaddlePaddle/PASSL/blob/main/docs/GETTING_STARTED.md) for the basic usage of PASSL.
2478

25-
## Tutorials
26-
- [Train SimCLR model](docs/Train_SimCLR_model.md)
27-
- [Train MoCo(v1,v2) model](docs/Train_MoCo_model.md)
28-
- [Train MoCo-BYOL model](docs/Train_MoCo-BYOL_model.md)
29-
- [Train BYOL model](docs/Train_BYOL_model.md)
30-
- [Train CLIP model](docs/Train_CLIP_model.md)
31-
- [Train PixPro model](docs/Train_PixPro_model.md)
79+
## Awesome SSL
80+
81+
Self-Supervised Learning (SSL) is a rapidly growing field, and some influential papers are listed here for research use.PASSL seeks to implement self-supervised algorithms with application potential
82+
83+
* *[Masked Feature Prediction for Self-Supervised Visual Pre-Training](https://arxiv.org/abs/2112.09133)* by Chen Wei, Haoqi Fan, Saining Xie, Chao-Yuan Wu, Alan Yuille, Christoph Feichtenhofer.
84+
* *[Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377)* by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
85+
* *[Corrupted Image Modeling for Self-Supervised Visual Pre-Training](https://arxiv.org/abs/2202.03382)* by Yuxin Fang, Li Dong, Hangbo Bao, Xinggang Wang, Furu Wei.
86+
* *[Are Large-scale Datasets Necessary for Self-Supervised Pre-training?](https://arxiv.org/abs/2112.10740)* by Alaaeldin El-Nouby, Gautier Izacard, Hugo Touvron, Ivan Laptev, Hervé Jegou, Edouard Grave.
87+
* *[PeCo: Perceptual Codebook for BERT Pre-training of Vision Transformers](https://arxiv.org/abs/2111.12710)* by Xiaoyi Dong, Jianmin Bao, Ting Zhang, Dongdong Chen, Weiming Zhang, Lu Yuan, Dong Chen, Fang Wen, Nenghai Yu.
88+
* *[SimMIM: A Simple Framework for Masked Image Modeling](https://arxiv.org/abs/2111.09886)* by Zhenda Xie, Zheng Zhang, Yue Cao, Yutong Lin, Jianmin Bao, Zhuliang Yao, Qi Dai, Han Hu.
89+
90+
## Contributing
91+
92+
PASSL is still young. It may contain bugs and issues. Please report them in our bug track system. Contributions are welcome. Besides, if you have any ideas about PASSL, please let us know.
93+
94+
## Citation
95+
96+
If PASSL is helpful to your research, feel free to cite
97+
98+
```
99+
@misc{=passl,
100+
title={PASSL: A visual Self-Supervised Learning Library},
101+
author={PASSL Contributors},
102+
howpublished = {\url{https://github.com/PaddlePaddle/PASSL}},
103+
year={2022}
104+
}
105+
```
106+
107+
## License
108+
109+
As shown in the LICENSE.txt file, PASSL uses the Apache 2.0 copyright agreement.

‎README_cn.md

+110
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,110 @@
1+
⚙️ 简体中文 | [English](./README.md)
2+
3+
<p align="center">
4+
<img src="./docs/imgs/passl_logo.svg" width="60%" height="60%"/>
5+
</p>
6+
<p align="center">
7+
<a href="./LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-red.svg"></a>
8+
<a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
9+
<a href="https://github.com/PaddlePaddle/PASSL/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PASSL?color=ccf"></a>
10+
<a href=""><img src="https://camo.githubusercontent.com/abb97269de2982c379cbc128bba93ba724d8822bfbe082737772bd4feb59cb54/68747470733a2f2f63646e2e7261776769742e636f6d2f73696e647265736f726875732f617765736f6d652f643733303566333864323966656437386661383536353265336136336531353464643865383832392f6d656469612f62616467652e737667"></a>
11+
<a href="https://aistudio.baidu.com/aistudio/personalcenter/thirdview/940489"><img src="https://img.shields.io/badge/Tutorial-AI Studio-blue.svg"></a>
12+
</p>
13+
14+
## 介绍
15+
16+
PASSL 是一个基于 PaddlePaddle 的视觉库，用于使用 PaddlePaddle 进行最先进的视觉自监督学习研究。PASSL旨在加速自监督学习的研究周期：**从设计一个新的自监督任务到评估所学的表征**。
17+
18+
PASSL 主要特性：
19+
20+
- 自监督前沿算法实现
21+
22+
PASSL 实现了多种前沿自监督学习算法，包括不限于 [SimCLR](https://arxiv.org/abs/2002.05709)、[MoCo(v1)](https://arxiv.org/abs/1911.05722)、[MoCo(v2)](https://arxiv.org/abs/1911.05722)、[MoCo-BYOL](docs/Train_MoCo-BYOL_model.md)、[CLIP](https://arxiv.org/abs/2103.00020)、[BYOL](https://arxiv.org/abs/2006.07733)、[BEiT](https://arxiv.org/abs/2106.08254)。同时支持有监督分类训练。
23+
24+
- 模块化设计
25+
26+
易于建立新的任务和重用其他任务的现有组件 (Trainer, models and heads, data transforms, etc.)
27+
28+
🛠️ PASSL 的最终目标是利用自监督学习为下游任务提供更合适的预训练权重，同时大幅度降低数据标注成本。
29+
30+
**📣 Recent Update:**
31+
32+
* (2022-2-9): 重构 README
33+
* 🔥 Now：PASSL 目前正在进行框架重构
34+
35+
## 模型库
36+
37+
* **Self-Supervised Learning Models**
38+
39+
PASSL 实现了一系列自监督学习算法，更具体的使用文档请参阅 **Document**
40+
41+
| | Epochs | Official results | PASSL results | Backbone | Model | Document |
42+
| --------- | ------ | ---------------- | ------------- | --------- | ------------------------------------------------------------ | ------------------------------------------------ |
43+
| MoCo | 200 | 60.6 | 60.64 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/moco_v1_r50_e200_ckpt.pdparams) | [Train MoCo](docs/Train_MoCo_model.md) |
44+
| SimCLR | 100 | 64.5 | 65.3 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/simclr_r50_ep100_ckpt.pdparams) | [Train SimCLR](docs/Train_SimCLR_model.md) |
45+
| MoCo v2 | 200 | 67.7 | 67.72 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/moco_v2_r50_e200_ckpt.pdparams) | [Train MoCo](docs/Train_MoCo_model.md) |
46+
| MoCo-BYOL | 300 | 71.56 | 72.10 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/mocobyol_r50_ep300_ckpt.pdparams) | [Train MoCo-BYOL](docs/Train_MoCo-BYOL_model.md) |
47+
| BYOL | 300 | 72.50 | 71.62 | ResNet-50 | [download](https://passl.bj.bcebos.com/models/byol_r50_300.pdparams) | [Train BYOL](docs/Train_BYOL_model.md) |
48+
| PixPro | 100 | 55.1(fp16) | 57.2(fp32) | ResNet-50 | [download](https://passl.bj.bcebos.com/models/pixpro_r50_ep100_no_instance_with_linear.pdparams) | [Train PixPro](docs/Train_PixPro_model.md) |
49+
50+
> Benchmark Linear Image Classification on ImageNet-1K.
51+
52+
Comming Soon：更多的算法实现已经在我们的计划中 ...
53+
54+
* **Classification Models**
55+
56+
PASSL 实现了视觉 Transformer 等具有影响力的图像分类算法，并提供了相应的预训练权重。旨在支持自监督、多模态、大模型算法的建设和研究。更多使用细节请参阅 [Classification_Models_Guide.md](docs/Classification_Models_Guide.md)
57+
58+
| | Detail | Tutorial |
59+
| ---------------- | --------------------------- | ------------------------------------------------------------ |
60+
| ViT | / | [PaddleEdu](https://aistudio.baidu.com/aistudio/projectdetail/2293050) |
61+
| Swin Transformer | / | [PaddleEdu](https://aistudio.baidu.com/aistudio/projectdetail/2280436) |
62+
| CaiT | [config](configs/cait) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401469) |
63+
| T2T-ViT | [config](configs/t2t_vit) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401348) |
64+
| CvT | [config](configs/cvt) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401386) |
65+
| BEiT | [config](configs/beit) | [unofficial](https://aistudio.baidu.com/aistudio/projectdetail/2417241) |
66+
| MLP-Mixer | [config](configs/mlp_mixer) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3401295) |
67+
| ConvNeXt | [config](configs/convnext) | [PaddleFleet](https://aistudio.baidu.com/aistudio/projectdetail/3407445) |
68+
69+
🔥 PASSL 提供了详细的算法剖析，具体请参阅 **Tutorial**。
70+
71+
## 安装
72+
73+
请参阅 [INSTALL.md](https://github.com/PaddlePaddle/PASSL/blob/main/docs/INSTALL.md) 进行安装
74+
75+
## 快速开始
76+
77+
请参阅 [GETTING_STARTED.md](https://github.com/PaddlePaddle/PASSL/blob/main/docs/GETTING_STARTED.md) 了解 PASSL 的基本用法
78+
79+
## Awesome SSL
80+
81+
自监督学习 (Self-Supervised Learning, SSL) 是一个发展十分迅速的领域，这里列出一些具有影响力的 Paper 供研究使用。PASSL 会争取实现具有应用潜力的自监督算法
82+
83+
* *[Masked Feature Prediction for Self-Supervised Visual Pre-Training](https://arxiv.org/abs/2112.09133)* by Chen Wei, Haoqi Fan, Saining Xie, Chao-Yuan Wu, Alan Yuille, Christoph Feichtenhofer.
84+
* *[Masked Autoencoders Are Scalable Vision Learners](https://arxiv.org/abs/2111.06377)* by Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, Ross Girshick.
85+
* *[Corrupted Image Modeling for Self-Supervised Visual Pre-Training](https://arxiv.org/abs/2202.03382)* by Yuxin Fang, Li Dong, Hangbo Bao, Xinggang Wang, Furu Wei.
86+
* *[Are Large-scale Datasets Necessary for Self-Supervised Pre-training?](https://arxiv.org/abs/2112.10740)* by Alaaeldin El-Nouby, Gautier Izacard, Hugo Touvron, Ivan Laptev, Hervé Jegou, Edouard Grave.
87+
* *[PeCo: Perceptual Codebook for BERT Pre-training of Vision Transformers](https://arxiv.org/abs/2111.12710)* by Xiaoyi Dong, Jianmin Bao, Ting Zhang, Dongdong Chen, Weiming Zhang, Lu Yuan, Dong Chen, Fang Wen, Nenghai Yu.
88+
* *[SimMIM: A Simple Framework for Masked Image Modeling](https://arxiv.org/abs/2111.09886)* by Zhenda Xie, Zheng Zhang, Yue Cao, Yutong Lin, Jianmin Bao, Zhuliang Yao, Qi Dai, Han Hu.
89+
90+
## 贡献
91+
92+
PASSL 还很年轻，它可能存在错误和问题。请在我们的错误跟踪系统中报告它们。我们欢迎您为 PASSL 做出贡献。此外，如果您对 PASSL 有任何想法，请告诉我们。
93+
94+
## 引用
95+
96+
如果 PASSL 对您的研究有帮助，欢迎引用
97+
98+
```
99+
@misc{=passl,
100+
title={PASSL: A visual Self-Supervised Learning Library},
101+
author={PASSL Contributors},
102+
howpublished = {\url{https://github.com/PaddlePaddle/PASSL}},
103+
year={2022}
104+
}
105+
```
106+
107+
## 开源许可证
108+
109+
如 LICENSE.txt 文件中所示，PASSL 使用 Apache 2.0 版权协议。
110+

‎docs/Usage_ViTs.md renamed to ‎docs/Classification_Models_Guide.md

File renamed without changes.

‎docs/imgs/passl_logo.svg

+9
Loading

0 commit comments

Comments
 (0)