PyTorch implementation of "Heterogeneous Graph Transformer for Multiple Tiny Object Tracking in RGB-T Videos", IEEE Transactions on MultiMedia. Feel free to contact me ([email protected]) if any questions. Please star the repository~~
- Download the dataset from Baidu Drive (Key: VTMT) and unzip them to
./data
- The multiple tiny object tracking dataset is composed of two modalities: visible and thermal, and is well aligned.
- Create the working environment through environment.yml
- Download the transformer pvtv2 backbone from PVTv2.
- Run training/main_RGBT-Tiny_graph_gnnloss.py for training.
- Run tracking/RGBT-Tiny_private_graph_Track2_crossmodal.py for tracking.
- This repository benefits a lot from TransCenter and GSDT.
@article{xu2024heterogeneous,
title={Heterogeneous Graph Transformer for Multiple Tiny Object Tracking in RGB-T Videos},
author={Xu, Qingyu and Wang, Longguang and Sheng, Weidong and Wang, Yingqian and Xiao, Chao and Ma, Chao and An, Wei},
journal={IEEE Transactions on Multimedia},
year={2024},
publisher={IEEE}
}