Cross-Camera Feature Prediction (CCFP)

The official implementation for the Cross-Camera Feature Prediction for Intra-Camera Supervised Person Re-identification across Distant Scenes which is accepted by ACMMM-2021.

Environment

The code is based on fastreid. See INSTALL.md.

For Compiling with cython to accelerate evalution

cd fastreid/evaluation/rank_cylib; make all

Dataset Preparation

  1. Download Market-1501 and DukeMTMC-reID
  2. Split Market-1501 and DukeMTMC-reID to Market-sct and DukeMTMC-sct according to the file names in the market_sct.txt and duke_sct.txt
  3. vim fastreid/data/build.py change the _root to your own data folder
  4. Make new directories in data and organize them as follows:
+-- data
|   +-- market
|       +-- market_sct
|       +-- query
|       +-- boudning_box_test
|   +-- duke
|       +-- duke_sct
|       +-- query
|       +-- boudning_box_test

Train and test

To train market-sct with CCFP, simply run

sh run.sh

To train duke-sct with CCFP, simply run

sh run_d.sh

Experiments

image

Citation

If you find this code useful, please kindly cite the following paper:

@inproceedings{ge2021cross,
  title={Cross-Camera Feature Prediction for Intra-Camera Supervised Person Re-identification across Distant Scenes},
  author={Ge, Wenhang and Pan, Chunyan and Wu, Ancong and Zheng, Hongwei and Zheng, Wei-Shi},
  booktitle={Proceedings of the 29th ACM International Conference on Multimedia},
  pages={3644--3653},
  year={2021}
}