dongb5 / adversarialbinarycoding4reid Goto Github PK

This project contains the codes for the paper Adversarial Binary Coding for Efficient Person Re-identification. We adopt adversarial learning to obtain compact discriminative binary representation for pedestrians and use them to measure similarity in an unified deep learning framework. The method is tested on CUHK03, Market-1501, and DukeMTMC-reID datasets.

The codes are tested in the Anaconda 4.4.0 environment containing following packages:

• Python 3.6
• PyTorch 0.3.0 + torchvision 0.12

• Train and test.

python main.py --dataset market1501 --train --test --trial 1 --data_dir /root/to/data

• draw.py for drawing loss curves.

By default, the '--data_dir' directs the 'data' folder in the root of the codes. Users can put the data folders into the 'data' folder. For convenience, we have put the train/test splitting file in pickle-readable format into the cuhk03 data folder.

@article{liu2018adversarial,
title={Adversarial Binary Coding for Efficient Person Re-identification},
author={Liu, Zheng and Qin, Jie and Li, Annan and Wang, Yunhong and Van Gool, Luc},
journal={arXiv preprint arXiv:1803.10914},
year={2018}
}