Unsupervised model for person Re Identification throught multiple camera
This is repository is largerly inspired by "Unsupervised Pre-training for Person Re-identification (2021)"[1] paper.
It aims to re-identify a person or a car through multiple camera thanks to an unsupervised siamese model trained on Market-1501[2]
The pipeline is composed with 4 main steps:
- Person detection using YoloV8 and image extraction of the interesting person
- Person detection on another video using YoloV8
- Tracking using StrongSORT
- Similarity between the reference and every person detected in the other video
Clone mikel-brostrom/yolov8_tracking repository
git clone https://github.com/mikel-brostrom/yolov8_tracking
To download the training dataset (Market-1501), use
curl https://www.kaggle.com/datasets/pengcw1/market-1501/download?datasetVersionNumber=1To download pretrained weights : https://kaiyangzhou.github.io/deep-person-reid/MODEL_ZOO
Then change
torchereidModels(name='<name>', weights='<path>')with your desired name and path
[1]
Unsupervised Pre-training for Person Re-identification (2021)
Dengpan Fu, Dongdong Chen, Jianmin Bao, Hao Yang, Lu Yuan, Lei Zhang, Houqiang Li, Dong Chen
arXiv:2012.03753
[2]
Market-1501
https://www.kaggle.com/datasets/pengcw1/market-1501
[3]
yolov8_tracking
mikel-brostrom
https://github.com/mikel-brostrom/yolov8_tracking
[4]
StrongSORT: Make DeepSORT Great Again
Yunhao Du, Yang Song, Bo Yang, and Yanyun Zhao
arXiv:2202.13514
[5]
deep-person-reid
Kaiyang Zhou
https://github.com/KaiyangZhou/deep-person-reid
[6]
Multi-Camera-Person-Tracking-and-Re-Identification
samihormi
https://github.com/samihormi/Multi-Camera-Person-Tracking-and-Re-Identification