reid

CAL-ReID

This folder contains the implementation of the person re-identification experiments.

Our implementation is based on the code of BoT.

Requirements

• Python 3.6+
• PyTorch 1.4
• torchvision
• CUDA 10.1
• ignite=0.1.2(https://github.com/pytorch/ignite)
• yacs(https://github.com/rbgirshick/yacs)

Prepare the data

• Download the Market1501, DukeMTMC-reID, and MSMT17 datasets.
• Change the path in \data\datasets\dataset_name.py to your dataset path.
• Change the path in \configs\softmax_triplet.yml to your model path.

Reproduce the results

• Run script train.sh to train the model.
• Change the path in test.sh to the pre-trained model path and run sh test.sh for inference.