Improving Pedestrian Attribute Recognition With Weakly-Supervised Multi-Scale Attribute-Specific Localization

Code for the paper "Improving Pedestrian Attribute Recognition With Weakly-Supervised Multi-Scale Attribute-Specific Localization", ICCV 2019, Seoul.

[Paper] [Poster]

Contact: [email protected] or [email protected]

Environment

• Python 3.6+
• PyTorch 0.4+

Datasets

• RAP: http://rap.idealtest.org/
• PETA: http://mmlab.ie.cuhk.edu.hk/projects/PETA.html
• PA-100K: https://github.com/xh-liu/HydraPlus-Net

The original datasets should be processed to match the DataLoader.

We provide the label lists for training and testing.

Training and Testing

python main.py --approach=inception_iccv --experiment=rap

python main.py --approach=inception_iccv --experiment=rap -e --resume='model_path'

Pretrained Models

We provide the pretrained models for reference, the results may slightly different with the values reported in our paper.

Dataset	mA	Link
PETA	86.34	Model
RAP	81.86	Model
PA-100K	80.45	Model

Reference

If this work is useful to your research, please cite:

@inproceedings{tang2019improving,
  title={Improving Pedestrian Attribute Recognition With Weakly-Supervised Multi-Scale Attribute-Specific Localization},
  author={Tang, Chufeng and Sheng, Lu and Zhang, Zhaoxiang and Hu, Xiaolin},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={4997--5006},
  year={2019}
}