jianwang-mpi / TextureGeneration

Pytorch implementation of Re-Identification Supervised Texture Generation

Re-ID Supervised Texture Generation

This is the code repo for the paper Re-Identification Supervised Texture Generation (CVPR2019) [PDF].

Requirement

• Python 3.6
• Pytorch 0.4.1

Install other python packages via:

pip install -r requirements.txt

Demo

• Download the pretrained weight

• Set the "model_path" in demo.sh to the path of pretrained weight

• Put some pedestrian images to example_results/input

• Run demo.sh

bash demo.sh

• Get the resulting textures from example_results/texture

• Render the 3D human model with texture using: BlenderRender

Train

1. Download datasets:

   • market-1501
   • SURREAL
   • CUHK-SYSU (for background)
   • PRW (for background)

2. Generate the rendering tensors with RenderingTensorGenerator.

3. Get the pretrained re-id network.

4. Set all paths and parameters in config.py

5. start train

bash train.sh

4. you will get the trained models in model_log_path

Citation

---

If you use this code for your research, please cite our paper.

@article{wang2019reidsupervised,
  title={Re-Identification Supervised Texture Generation},
  author={Jian, Wang and Yunshan, Zhong and Yachun, Li and Chi, Zhang and Yichen, Wei},
  journal={IEEE Conference on Computer Vision and Pattern Recognition},
  year={2019}
}