This repo contains the code for our paper A Simple Video Segmenter by Tracking Objects Along Axial Trajectories.
Note: We update our model name from MaXTron to Axial-VS in the TMLR version to better reflect the key innovations in our paper. To keep minimal change in the released codebase, we still maintain the usage of MaXTron below.
MaXTron is a simple yet effective unified meta-architecture for video segmentation, which enriches existing clip-level segmenters by introducing a within-clip tracking module and a cross-clip tracking module, thus achieving better temporally consistent segmentation results.
For detailed usage of MaXTron, see
If you use MaXTron (Axial-VS) in your research, please use the following BibTeX entry.
@article{
he2024a,
title={A Simple Video Segmenter by Tracking Objects Along Axial Trajectories},
author={Ju He and Qihang Yu and Inkyu Shin and Xueqing Deng and Alan Yuille and Xiaohui Shen and Liang-Chieh Chen},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2024},
url={https://openreview.net/forum?id=Sy6ZOStz5v},
note={}
}