«RethinkPAR» provides a StrongBaseline for Pedestrian Attribute Recognition

• Table of Contents
• Latest News
• Background
• Installation
• Dataset
• Usage
  • Train
  • Eval
• Maintainers
• Thanks
• Contributing
• License

• [2023/11/09]v0.1.2. Update ResNet101 training result.
• [2023/11/08]v0.1.1. Update BCELoss and Train settings.
• [2023/11/07]v0.1.0. Baseline(ResNet50) + PETA_zs.

The paper Rethinking of Pedestrian Attribute Recognition: A Reliable Evaluation under Zero-Shot Pedestrian Identity Setting provides a detailed definition of existing research in the field of pedestrian attribute recognition, not only providing a clear definition of pedestrian attribute recognition, but also providing a new baseline method.

I tried the code repository provided in the paper, but there were several obvious issues, such as key dependency libraries not adapting to the latest version (unable to use Pytorch 2. x.x), and the overall implementation of the project being too heavy for further development and integration.

In order to facilitate better research and application of the methods proposed in this paper, I have implemented this warehouse to simplify training, evaluation, and prediction as much as possible.

pip install -r requirements.txt

The application for the RAP dataset is very complicated and requires filling in relevant records. So this warehouse only uses the PETA dataset for experiments

PETA_zs used 35 out of the 105 attributes of the original dataset PETA, as detailed in: PETA （35 in 105）

• Download link:Baidu Drive
• Password: nktm

CUDA_VISIBLE_DEVICES=0 python train.py ../datasets/PETA/ runs/r50_train_b64/ --backbone resnet50 --num_attr 32

python eval.py ../datasets/PETA/ runs/r50_train_b64/rethinking_par-e95.pth 

• zhujian - Initial work - zjykzj

• valencebond/Rethinking_of_PAR
• zjykzj/crnn-ctc

Anyone's participation is welcome! Open an issue or submit PRs.

Small note:

• Git submission specifications should be complied with Conventional Commits
• If versioned, please conform to the Semantic Versioning 2.0.0 specification
• If editing the README, please conform to the standard-readme specification.

Apache License 2.0 © 2023 zjykzj