metadata
tags:
- pytorch
- classification
pipeline_tag: image-segmentation
library_name: pytorch
license: mit
MoSeg Model: Segment Any Motion in Videos
This model, presented in the paper Segment Any Motion in Videos, performs moving object segmentation in videos. Unlike methods relying solely on optical flow, MoSeg combines long-range trajectory motion cues with DINO-based semantic features and leverages SAM2 for precise pixel-level mask generation. It excels in challenging scenarios with complex deformations and background distractions, achieving state-of-the-art performance on diverse datasets.
Code: https://github.com/nnanhuang/SegAnyMo
Project Page: https://motion-seg.github.io/