[ICASSP] Graph Convolution for Re-ranking in Person Re-identification

The official repository for GCR rerank, a GCN-based reranking method for both image and video re-ID.


We use python 3.7/torch 1.6/torchvision 0.7.0.

Extracted features

We provide Market1501/MARS features from reid-strong-baseline at Google Drive.

Command Lines

Run GCRV rerank with basic settings on Market1501

python eval_rerank.py --config_file=config/market.yml

Run PVG only

python eval_rerank.py --config_file=config/market.yml PVG.ENABLE_PVG True GCR.ENABLE_GCR False

Run GCR only

python eval_rerank.py --config_file=config/market.yml PVG.ENABLE_PVG False GCR.ENABLE_GCR True

RUN GCRV on video reid dataset(MARS)

python eval_rerank.py --config_file=config/mars.yml

Run other rerank methods: (baseline, k_reciprocal, ecn, ecn_orig, lbr, qe)

python eval_rerank.py --config_file=config/market.yml COMMON.RERANK_TYPE baseline


State-of-the-art reranking method inlucidng K_reciprocal, ECN, LBR


If you find our work useful in your research, please consider citing:

 title={Graph Convolution for Re-ranking in Person Re-identification},
 author={Zhang, Yuqi and Qian Qi and Liu Chong and Chen, Weihua and Wang Fan and Li Hao and Jin Rong},
 journal={arXiv preprint arXiv:2107.02220},