MMsAGN

Masked Graph Structure and Multi-scale Attention-based Graph Network for Vehicle Re-Identification

Train

Training on VeRi-776 dataset：

python tools/train_net.py --config-file ./configs/VeRi/sbs_R50-ibn.yml

Training on VehicleID dataset：

python tools/train_net.py --config-file ./configs/VehicleID/Base-bagtricks.yml

Training with four GPUs：

python3 tools/train_net.py --config-file ./configs/VeRi/sbs_R50-ibn.yml --num-gpus 4

Test:

python tools/train_net.py --config-file ./configs/VeRi/sbs_R50-ibn.yml --eval-only MODEL.WEIGHTS ./logs/veri776/model_best.pth

You can use our trained weight files for testing. You can access the files through this link.

Result

VeRi-776

Performance (%) comparison on the VeRi-776 dataset.

Method	mAP	Rank-1	Rank-5
Baseline	81.09	96.72	98.33
MMsAGN	83.57	97.44	99.05

VehicleID

Performance (%) comparison on the VehicleID dataset.

Method	Test800 Rank-1	Test800 Rank-5	Test1600 Rank-1	Test1600 Rank-5	Test2400 Rank-1	Test2400 Rank-5
Baseline	66.33	89.38	58.72	82.55	53.77	76.83
MMsAGN	87.41	98.17	83.88	96.36	81.30	94.22