Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/ezra-yu/accv2022_fgia_1st

1st solution for the Webly-supervised Fine-grained Recognition competition in https://www.cvmart.net/race/10412/base
https://github.com/ezra-yu/accv2022_fgia_1st

arcface-loss eql fine-grained-classification longtail mae pseudolabelling pytorch swin

Last synced: 5 days ago
JSON representation

1st solution for the Webly-supervised Fine-grained Recognition competition in https://www.cvmart.net/race/10412/base

Host: GitHub
URL: https://github.com/ezra-yu/accv2022_fgia_1st
Owner: Ezra-Yu
Created: 2022-09-29T08:16:46.000Z (over 2 years ago)
Default Branch: master
Last Pushed: 2023-03-08T03:06:21.000Z (almost 2 years ago)
Last Synced: 2024-12-29T13:44:49.786Z (12 days ago)
Topics: arcface-loss, eql, fine-grained-classification, longtail, mae, pseudolabelling, pytorch, swin
Language: Python
Homepage:
Size: 34.7 MB
Stars: 33
Watchers: 3
Forks: 1
Open Issues: 0
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

        # Solution of FGIA ACCV 2022(1st Place)

This is the 1st Place Solution for Webly-supervised Fine-grained Recognition, refer to the ACCV workshop competition in https://www.cvmart.net/race/10412/base.

Mainly done by [Ezra-Yu](https://github.com/Ezra-Yu), [Yuan Liu](https://github.com/YuanLiuuuuuu) and [Songyang Zhang](https://github.com/tonysy), base on [**MMClassifiion**](https://github.com/open-mmlab/mmclassification) 与 [**MMSelfSup**](https://github.com/open-mmlab/mmselfsup). please flork and star them if you think they are useful.

**[Lecture](https://www.bilibili.com/video/BV13M411S7b5/?spm_id_from=333.337.search-card.all.click)** and **[slide](https://github.com/Ezra-Yu/ACCV_STORE/releases/download/v0.0.1/ACCV_FGIA_1st_PPT.pdf)** are available.

#### Note:

Tricks ([`AdaptiveSubCenterArcFace`](https://github.com/open-mmlab/mmclassification/tree/dev-1.x/configs/arcface), `EQL_Loss`, `Post-Hoc_LT_Adjusment`, `Uniform_Model_Soup`) have been or will be implemented in [MMClassifiion](https://github.com/open-mmlab/mmclassification).

## Result

Show the result




**LB A**

![LB-A](https://user-images.githubusercontent.com/18586273/205498131-5728e470-b4f6-43b7-82a5-5f8e3bd5168e.png)

**LB B**

![LB-B](https://user-images.githubusercontent.com/18586273/205498171-5a3a3055-370a-4a8b-9779-b686254ebc94.png)

## Reproduce / 复现

**复现精度请点击[这里](./Reproduce.md)**

## Description

### Backbone and Pre-train

- ViT(MAE-pt)   # 自己预训练

- Swin(21kpt)   # 来自[MMCls-swin_transformer_v2](https://github.com/open-mmlab/mmclassification/tree/dev-1.x/configs/swin_transformer_v2).

**Archs**

- ViT + CE-loss + post-LongTail-Adjusment                 

- ViT + SubCenterArcFaceWithAdvMargin(CE)              

- Swin-B + SubCenterArcFaceWithAdvMargin(SoftMax-EQL)  

- Swin-L + SubCenterArcFaceWithAdvMargin(SoftMAx-EQL) 

所有都使用了 **Flip TTA**。

## Flow

MIM 预训练 --> 训练 --> 清洗数据 --> fine-tune + 集成 + 生成伪标签交替训练 --> 后处理

1. MAE 预训练

2. Swin 与 ViT 的训练

3. 使用权重做数据清洗         

4. 训练 -> 制作伪标签，放回训练集中 -> 再训练； (testa3轮)

5. 训练 -> 制作伪标签，放回训练集中 -> 再训练； (testb3轮,包括testa的伪标签)

6. 模型融合，调整预测的标签分布，提交

![flow](https://user-images.githubusercontent.com/18586273/205498371-31dbc1f4-5814-44bc-904a-f0d32515c7dd.png)

## Summary

- [MAE](https://github.com/open-mmlab/mmselfsup/tree/dev-1.x/configs/selfsup/mae) |  [Config](./configs/vit/)  

- [Swinv2](https://github.com/open-mmlab/mmclassification/tree/dev-1.x/configs/swin_transformer_v2) | [Config](./configs/swin/) 

- [ArcFace](https://arxiv.org/abs/1801.07698)   |   [Code](./src/models/arcface_head.py)  

- [SubCenterArcFaceWithAdvMargin](https://paperswithcode.com/paper/sub-center-arcface-boosting-face-recognition)   |   [Code](./src/models/arcface_head.py) 

- [Post-LT-adjusment](https://paperswithcode.com/paper/long-tail-learning-via-logit-adjustment)   |   [Code](./src/models/linear_head_lt.py) 

- [SoftMaxEQL](https://paperswithcode.com/paper/the-equalization-losses-gradient-driven)   |   [Code](./src/models/eql.py)   

- FlipTTA  |   [Code](./src/models/tta_classifier.py)   

- 数据清洗                                                

- 模型融合: [Uniform-model-soup](https://arxiv.org/abs/2203.05482) | [code](./tools/model_soup.py)            

- [半监督](https://lilianweng.github.io/posts/2021-12-05-semi-supervised/)  | [Code](./tools/creat_pseudo.py)  

- 自适应集成 [Code](./tools/emsemble.py),                  

- 后处理: [调整预测分布](./tools/re-distribute-label.py);    

![image](https://user-images.githubusercontent.com/18586273/205498027-def99b0d-a99a-470b-b292-8d5fc83111fc.png)

#### 使用了没有效果 used but no improvement

1. 使用检索以及检索分类混合的方式

2. 使用EfficientNet

#### 后续 not used but worth try

1. DiVE 蒸馏提升长尾问题表现

2. Simim 训一个 swinv2 的预训练模型

3. 优化 re-distribute-label