- Pytorch 1.1
- cuda 9.0
- python 3.6
- GPU Memory>20G We Recommend Titan RTX or Tesla V100
We evaluate our method on Mars, iLIDS-VID and PRID-2011 datasets. You can download datasets from Here, and put them into /data/datasets/.
- Firstly, we provide the region box information which extracts from Mars,iLIDS-VID and PRID-2011 datasets with the application of HRNet. You can download from the following links and put them into /data/keypoints/.
MARS_Testing_RegionBox| MARS_Training_RegionBox| iLIDS-VID Testing and Training RegionBox| PRID-2011 Testing and Training RegionBox
- If you want to test our trained model on MARS, you can obtain our trained model from Here, and put it into /weights/. After that, you can run our code with the following command:
python evaluate.py --dataset mars
- If you want to train the network, you can run our code with the following commands:
On Mars dataset:
python run.py --dataset mars --max-epoch 400 --train-batch 32 --num-instances 4 --lr 0.00035 --feat-dim 1024 --a1 1 --a2 1 --a3 0.0003 --margin 0.5 --gpu-devices 0
On iLIDS-VID dataset:
python run.py --dataset ilidsvid --max-epoch 400 --train-batch 32 --num-instances 4 --lr 0.00035 --feat-dim 512 --a1 1 --a2 1 --a3 0.00005 --margin 0.5 --gpu-devices 0
On PRID-2011 dataset:
python run.py --dataset prid --max-epoch 400 --train-batch 32 --num-instances 4 --lr 0.00035 --feat-dim 256 --a1 1 --a2 1 --a3 0.00005 --margin 0.5 --gpu-devices 0
Dataset | Rank1 | Rank5 | Rank20 | mAP |
---|---|---|---|---|
MARS | 88.7 | 96.1 | 98.1 | 82.8 |
PRID2011 | 95.2 | 99.2 | 100.0 | - |
iLIDS-VID | 89.6 | 98.0 | 99.9 | - |