What I did

I visualize the result of "spatial transformer network" module of "Improving Pedestrian Attribute Recognition With Weakly-Supervised Multi-Scale Attribute-Specific Localization". I follows visualization method of original paper.

see model/inception.py -> class inceptionnet -> def visualization

How to visualize?

• First download rap directory (root 디렉토리에 rap dataset 위치시킴)
• make named "visualization" empty directory in project directory. (root 디렉토리에 visualization이란 이름의 디렉토리 생성)
• After finishing training, you can make visualized input pictures by giving "--visualization=true" option. (training 이후에 model이 'saved_model.pt'라는 이름으로 root 디렉토리에 자동 저장됨. visualization 시에는 저장된 모델을 불러와서 시각화된 이미지파일들을 visualization 디렉토리에 저장함)

Visualization=True option may fill your empty named "visualization" directory

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

Below is the description of original code of ICCV19, Improving pedestrian Attribute Recognition with Weakly-Supervised Multi-Scale Attribute-Specific-Localization" Yu can find original code in here (https://github.com/chufengt/iccv19_attribute)

---

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}
}