Multi-Granularity Video Object Segmentation

Sangbeom Lim^1* · Seongchan Kim^1* · Seungjun An^3* · Seokju Cho² · Paul Hongsuck Seo^1† . Seungryong Kim^2†

¹Korea University    ²KAIST    ³Samgsung Electronics

*: Equal Contribution
†: Co-Corresponding Author

AAAI 2025

MUG-VOS contains multiple granularities masks from coarse to fine segments.

📰 News

• 2024-12-01: MUG-VOS is released.
• 2024-12-20: Training Code, Data collection pipeline, and MUG-VOS Testset are released.

Please stay tuned for a MUG-VOS v2!

Evaluation Dataset Preparation

First, download the evaluation datasets:

• Huggingface Dataset

or

git clone https://huggingface.co/datasets/SammyLim/MUG-VOS

Data Collection Pipeline

For detailed instructions on Data Collection Pipeline, please refer to the README file for your chosen implementation:

• PyTorch Implementation

Training and Evaluation

For detailed instructions on training and evaluation, please refer to the README file for your chosen implementation:

• PyTorch Implementation

📚 Citing this Work

Please use the following bibtex to cite our work:

@article{lim2024multi,
  title={Multi-Granularity Video Object Segmentation},
  author={Lim, Sangbeom and Kim, Seongchan and An, Seungjun and Cho, Seokju and Seo, Paul Hongsuck and Kim, Seungryong},
  journal={arXiv preprint arXiv:2412.01471},
  year={2024}
}

🙏 Acknowledgement

This project is largely based on the XMem repository. Thanks to the authors for their invaluable work and contributions.