We follow the IRRA dataloader code. Just set the dataset path in configs PATH.DATASETS option.
Use the vscode launch file or use the following bash command line:
python train.py --config-file configs/cuhk_pedes/two_stream.yaml --device-num 0On CUHK-PEDES
| Method | Rank-1 | Rank-5 | Rank-10 | mAp | mINP |
|---|---|---|---|---|---|
| LFSA | 75.20 | 89.23 | 93.41 | 69.50 | 56.00 |
On ICFG-PEDES
| Method | Rank-1 | Rank-5 | Rank-10 | mAp | mINP |
|---|---|---|---|---|---|
| LFSA | 66.83 | 80.07 | 84.88 | 46.85 | 14.28 |
The code structure is based on TextReID, the dataloader and optimization part refers the IRRA, and the model code refers the CLIP. Many thanks to them for their contributions in this field.