/CDL

Code for our paper: Consistent Discrepancy Learning for Intra-camera Supervised Perosn Re-Identification

Primary LanguagePythonMIT LicenseMIT

CDL

Official PyTorch implementation of CDL: Consistent Discrepancy Learning for Intra-camera Supervised Perosn Re-Identification.

Change the dataset path in lines 450-455 in utils.py to your path to the dataset. Set the ``load_weight'' in exp_1.yaml to your path to the pretrained model. The pretrained model is trained to classify the persons under each view and learn pedestrian prototypes[1]. We provide our pretrained model at:https://pan.baidu.com/s/1dduMG0FEvWSSHWpei9v_sQ?pwd=cxub key: 【cxub】

Example: python main.py --gpu 0,1 --config config/exp_1.yaml

This project is implemented based on:

  1. https://github.com/KovenYu/MAR
  2. https://github.com/KaiyangZhou/deep-person-reid

[1]. NormFace: L2 Hypersphere Embedding for Face Verification