Graph Neural Networks for Cross-Camera Data Association

IEEE TCSVT Paper: https://ieeexplore.ieee.org/document/9893862

Setup & Running

Requirements

The repository has been tested in the following software.

• Ubuntu 16.04
• Python 3.6
• Anaconda
• Pycharm

1. Clone repository

git clone https://github.com/vpulab/GNN-Cross-Camera-Association.git

2. Anaconda environment

To create and setup the Anaconda Envirmorent run the following terminal command from the repository folder:

conda env create -f env_gnn.yml
conda activate env_gnn

3. Install Torchreid library

git clone https://github.com/KaiyangZhou/deep-person-reid.git
cd deep-person-reid/
python setup.py develop
cd ..

4. Download and prepare EPFL dataset

This repo is evaluated on EPFL Terrace (seq. 1), Laboratory (seq. 6p), and Basketball sequence.

4a. To automatically download the sequences run

download_dataset.sh

or,

4b. To do it by your own download the EPFL video sequences at https://www.epfl.ch/labs/cvlab/data/data-pom-index-php/. Then, place each .avi sequence in their corresponding path, e.g. ./datasets/EPFL-Terrace/terrace1-c0/terrace1-c0.avi and name each .avi as the name of the folder containing it.

5. Run

python ./libs/preprocess_EPFL.py

in order to extract frame images.

6. Ground-truth

The EPFL GT (we already provide it, no need to download it) can be found at https://bitbucket.org/merayxu/multiview-object-tracking-dataset/src/master/.

7. Download pre-trained REID models

Download the pre-trained REID models from here , unzip the 4 folders and place them under ./trained_models/

8. Download a pre-trained GNN-CCA model

We provide the weights of the GNN trained on the S1 set (see paper for detailes). Download the pre-trained weights from here and place the folder GNN_S1_Resnet50MCD_SGD0005_cosine20_BS64_BCE_all_step_BNcls_L4_2021-11-10 19:01:49 under ./results/ folder.

9. Inference Running

To inference the previous model run:

python main.py --ConfigPath "config/config_inference.yaml"

10. Training

For training run:

python main_training.py --ConfigPath "config/config_training.yaml"

Citation

If you find this code and work useful, please consider citing:

@ARTICLE{9893862,
  author={Luna, Elena and SanMiguel, Juan C. and Martínez, José M. and Carballeira, Pablo},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Graph Neural Networks for Cross-Camera Data Association}, 
  year={2022},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TCSVT.2022.3207223}}
}