A simple tool based on SiamX for visual object tracking using normal or 360 images (euqirectangular projection). SiamX is an efficient long-term tracker achieving state-of-the-art results and runs at higher speed. If you are interested in the tracker SiamX, you could read the paper and find more detail in our project page. In this repository, it only exploits the network architecture of SiamX while does not contain the code for training and dataset evaluation.
cd $360Tracking
conda env create -f environment.yml
conda activate 360tracking
Please download the SiamX model here, and then uncompress and put it in ./SiamX.
In root path $360Tracking,
python tools/run.py --video YOUR_VIDEO_PATH --tracker omni --resume Your_Snapshot_Path
e.g., python tools/run.py --video ./SiamX/demo.mp4 --tracker omni --resume ./SiamX/snapshot/SiamX.pth
The default --tracker base does not support cross-boundary tracking. If the video is composed of 360 images, you should select --tracker omni for better performance.
You could also download and use our data for testing.
If you find any part of our work useful in your research, please consider citing our paper:
@inproceedings{hhuang2022siamx,
title = {SiamX: An Efficient Long-term Tracker Using Cross-level Feature Correlation and Adaptive Tracking Scheme},
author = {Huang, Huajian and Yeung, Sai-Kit},
booktitle = {International Conference on Robotics and Automation (ICRA)},
year = {2022},
organization={IEEE}
}