Introduction
This is the PyTorch implementation for Adaptive Graph Representation Learning for Video Person Re-identification.
Get started
git clone https://github.com/weleen/AGRL.pytorch /path/to/save
pip install -r requirements.txt
cd torchreid/metrics/rank_cylib && make
Dataset
create dataset directory
mkdir data
Prepare datasets:
├── dukemtmc-vidreid
│ ├── DukeMTMC-VideoReID
│ ├── pose.json
│ ├── split_gallery.json
│ ├── split_query.json
│ └── split_train.json
│
├── ilids-vid
│ ├── i-LIDS-VID
│ ├── pose.json
│ ├── splits.json
│ └── train-test people splits
│
├── mars
│ ├── bbox_test
│ ├── bbox_train
│ ├── info
│ ├── pose.json
│ └── train-test people splits
│
├── prid2011
├── pose.json
├── prid_2011
├── prid_2011.zip
├── splits_prid2011.json
└── train_test_splits_prid.mat
pose.json is obtained by running AlphaPose, we put the files on Baidu Netdisk (code: luxr) and Google Driver.
More details could be found in DATASETS.md.
Train
bash scripts/train_vidreid_xent_htri_vmgn_mars.sh
To use multiple GPUs, you can set --gpu-devices 0,1,2,3.
Note: To resume training, you can use --resume path/to/model to load a checkpoint from which saved model weights and start_epoch will be used. Learning rate needs to be initialized carefully. If you just wanna load a pretrained model by discarding layers that do not match in size (e.g. classification layer), use --load-weights path/to/model instead.
Please refer to the code for more details.
Test
create a directory to store model weights mkdir saved-models/ beforehand. Then, run the following command to test
bash scripts/test_vidreid_xent_htri_vmgn_mars.sh
All the model weights are available.
Model
All the results tested with 4 TITAN X GPU and 64GB memory.
| Dataset | Rank-1 | mAP |
|---|---|---|
| iLIDS-VID | 83.7% | - |
| PRID2011 | 93.1% | - |
| MARS | 89.8% | 81.1% |
| DukeMTMC-vidreid | 96.7% | 94.2% |
Citation
Please kindly cite this project in your paper if it is helpful
@article{wu2020adaptive,
title={Adaptive graph representation learning for video person re-identification},
author={Wu, Yiming and Bourahla, Omar El Farouk and Li, Xi* and Wu, Fei and Tian, Qi and Zhou, Xue},
journal={IEEE Transactions on Image Processing},
year={2020},
publisher={IEEE}
}
This project is developed based on deep-person-reid and STE-NVAN.
712 Dec 17, 2022
143 Dec 24, 2022
4 Jun 17, 2021
77 Jan 3, 2023
40 Nov 25, 2022
1.3k Jan 1, 2023
45 Dec 3, 2021
12 Dec 12, 2022
70 Oct 10, 2022
6 Nov 25, 2022
62 Dec 21, 2022
1 Apr 9, 2022
41 Jan 6, 2023
30 Jul 14, 2022
3.7k Jan 5, 2023
1.2k Dec 30, 2022
17 Dec 20, 2022
3 Jun 17, 2022