ICCV19 Pose-Guided Feature Alignment Occluded Person ReID

This is the pytorch implementation and dataset of the ICCV2019 paper "Pose-Guided Feature Alignment for Occluded Person Re-identification"

Preparation

Dependencies

• Python 3.7
• Pytorch 1.0
• Numpy

Occluded-DukeMTMC Dataset Preparation

The Occluded-DukeMTMC dataset is re-splited based on the DukeMTMC dataset. Since the privacy implications of the data set are being considered, we cannot release the images of Occluded-DukeMTMC. We only release the image name lists of our Occluded-DukeMTMC dataset in './dataset/Occluded_Duke'. If you can access to the DukeMTMC-reid dataset, you can easily convert DukeMTMC-reid to Occluded-DukeMTMC by running the following code

cd dataset
python convert_duke_to_occduke.py /path/to/DukeMTMC-reID.zip
cd ..

Pose landmarks extraction (optional)

Use AlphaPose to extract pose landmarks of the training set and testing set.

cd AlphaPose
sh infer.sh
cd ..

Or you can download our extracted pose landmarks and generated heatmaps.

Download pose landmarks and heatmaps

Download pose landmarks and heatmaps into the root path. Unzip them.

Train

python train.py

Test

python test.py

Inference on Partial_REID or Partial_iLIDS

./dataset2.zip : Partial_REID and Partial_iLIDS with holistic images.

You can directly use the processed data to run the code.

Data link: https://drive.google.com/file/d/1ErCEQsNHSHpgZF3-NNj6_OH322vpk8gn/view?usp=sharing

Model link: https://drive.google.com/file/d/1VarCCCaWZlDYX3La2r8VZpB9rZoHocMm/view?usp=sharing

Heatmaps link: https://drive.google.com/file/d/1VAmMgGym9XfxAMeq_YmzydDI50KTqnsl/view?usp=sharing

   GALLERY_DIR='/your/path/to/heatmaps/Partial_REID/18heatmap_gallery'
   QUERY_DIR='/your/path/to/heatmaps/Partial_REID/18heatmap_query'
   gallery_pose_dir='your/path/to/heatmaps/Partial_REID/gallery_json_1'
   query_pose_dir='your/path/to/heatmaps/Partial_REID/query_json_1'
   python test.py --name market_ckp --part_num 6 --test_dir /your/dataset/path/ —-gallery_heatmapdir $GALLERY_DIR --query_heatmapdir $QUERY_DIR --gallery_posedir $gallery_pose_dir --query_posedir $query_pose_dir --train_classnum 751

Citation

Please cite this paper in your publications if it helps your research:

@inproceedings{miao2019pose,
  title={Pose-guided feature alignment for occluded person re-identification},
  author={Miao, Jiaxu and Wu, Yu and Liu, Ping and Ding, Yuhang and Yang, Yi},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={542--551},
  year={2019}
}