Breaking the Positive Sample Barrier in Person Re-Identification: Towards Domain Generalization without Paired Samples
pip install -r requirement.txt
- requirement:
Python 3.9.0
Pytorch 1.10.0 & torchvision 0.11.0
- Download the datasets(Market-1501, MSMT17)and then unzip them to
your_dataset_dir
. - Split Market-1501 and MSMT to Market-SCT and MSMT-SCT according to CCFP.
- Make new directories in data and organize them as follows:
+-- data | +-- market1501 | +-- boudning_box_train | +-- query | +-- boudning_box_test | +-- market1501_sct | +-- boudning_box_train_sct | +-- query | +-- boudning_box_test | +-- MSMT17 | +-- train_sct | +-- test | +-- list_train.txt | +-- MSMT_mixSCT.txt
train
CUDA_VISIBLE_DEVICES=0 python train.py --config-file configs/Market_SCT/vit_transreid.yml
test
CUDA_VISIBLE_DEVICES=0 python test.py --config-file configs/Market_SCT/vit_transreid.yml
If you have any questions, please feel free to contact me(lyx520419@163.com).