- π We only trained one model (ResNeSt) with different scales (i.e., 224, 256, and 288), respectivel achieved 91.7% and 86.27% in phase A and B.
- π Traing time cost ~1.5 hour with a V100 16GB, so easy, no bells and whistles!
- π Techical details are described in our arXiv preprint paper.
- π The test data can be obtained from here.
- π The offline data can be obtained from here [d0kc].
- π The weights can be obtained from here [ba2y].
- Click on the star β, Thank you β€οΈ
- PyTorch 1.7.0+cu101
- torchvision 0.8.1+cu101
pip install -r requirements.txt; cd fastreid/evaluation/rank_cylib; make all
bash train1.sh
bash train2.sh
bash train3.sh
bash train4.sh
bash predict.sh
A large portion of code is borrowed from fast-reid, many thanks π to their wonderful work!