MOVE: Motion-Guided Few-Shot Video Object Segmentation

TL;DR

This paper introduces MOVE, a large-scale dataset for motion-guided few-shot video object segmentation, evaluates existing methods, identifies challenges, and proposes a baseline approach to improve motion understanding in videos.

Contribution

The paper presents MOVE, the first dataset specifically designed for motion-guided FSVOS, and introduces DMA, a baseline method that advances few-shot motion understanding.

Findings

Current methods struggle with motion-guided FSVOS.

MOVE dataset enables comprehensive evaluation of motion understanding.

DMA outperforms existing approaches in few-shot motion tasks.

Abstract

This work addresses motion-guided few-shot video object segmentation (FSVOS), which aims to segment dynamic objects in videos based on a few annotated examples with the same motion patterns. Existing FSVOS datasets and methods typically focus on object categories, which are static attributes that ignore the rich temporal dynamics in videos, limiting their application in scenarios requiring motion understanding. To fill this gap, we introduce MOVE, a large-scale dataset specifically designed for motion-guided FSVOS. Based on MOVE, we comprehensively evaluate 6 state-of-the-art methods from 3 different related tasks across 2 experimental settings. Our results reveal that current methods struggle to address motion-guided FSVOS, prompting us to analyze the associated challenges and propose a baseline method, Decoupled Motion Appearance Network (DMA). Experiments demonstrate that our approach achieves superior performance in few shot motion understanding, establishing a solid foundation for future research in this direction.