- Python 3.7
- PyTorch 1.10
- Ignite 0.4.7
- Yacs
Download SYSU-MM01 dataset and uncompress it.
Change the entry data_root
in configs/default.py to the path of the dataset.
Put the rand_perm_cam.mat in exp
directory in dataset root. This file is used to assign gallery items for each trial while testing.
Run
CUDA_VISIBLE_DEVICES=0 python3 train.py configs/baseline.yml --work-dir work_dirs/baseline
We evaluate the performance on SYSU-MM01 under the setting of one-shot & all-search.
model | mAP | rank-1 | rank-5 | rank-10 | rank-20 |
---|---|---|---|---|---|
baseline | 54.60 | 57.51 | 82.77 | 90.05 | 95.28 |