Skip to content

CVPR 2018 Exploit the Unknown Gradually: One-Shot Video-Based Person Re-Identification by Stepwise Learning

License

Notifications You must be signed in to change notification settings

Yu-Wu/Exploit-Unknown-Gradually

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

25 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Pytorch implementation for our paper [Link]. This code is based on the Open-ReID library.

Preparation

Dependencies

  • Python 3.6
  • PyTorch (version >= 0.2.0)
  • h5py, scikit-learn, metric-learn, tqdm

Download datasets

  • DukeMTMC-VideoReID: This page contains more details and baseline code.
  • MARS: [Google Drive]   [BaiduYun].
  • Move the downloaded zip files to ./data/ and unzip here.

Train

For the DukeMTMC-VideoReID dataset:

python3 run.py --dataset DukeMTMC-VideoReID --logs_dir logs/DukeMTMC-VideoReID_EF_10/ --EF 10 --mode Dissimilarity --max_frames 900

For the MARS datasaet:

python3 run.py --dataset mars --logs_dir logs/mars_EF_10/ --EF 10 --mode Dissimilarity --max_frames 900

It takes about 10 hours to train EUG (EF=10%) on DukeMTMC-VideoReID with a GTX1080Ti. Please set the max_frames smaller if your GPU memory is less than 11G.

Performances

The performances varies according to random splits for initial labeled data. To reproduce the performances in our paper, please use the one-shot splits at ./examples/

Citation

Please cite the following paper in your publications if it helps your research:

@inproceedings{wu2018cvpr_oneshot,
    title = {Exploit the Unknown Gradually: One-Shot Video-Based Person Re-Identification by Stepwise Learning},
    author = {Wu, Yu and Lin, Yutian and Dong, Xuanyi and Yan, Yan and Ouyang, Wanli and Yang, Yi},
    booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    year = {2018}
}

Future work

In the extended version, We improve EUG by leveraging the remaining unselected data in an unsupervised way, which shows a great performance improvement on the image-based re-ID datasets (Market-1501 and DukeMTMC-reID). [Paper] [Code]

Contact

To report issues for this code, please open an issue on the issues tracker.

If you have further questions about this paper, please do not hesitate to contact me.

Yu Wu's Homepage

About

CVPR 2018 Exploit the Unknown Gradually: One-Shot Video-Based Person Re-Identification by Stepwise Learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published