MASSeg : 2nd Technical Report for 4th PVUW MOSE Track

2025-04-14

AI Summary

Addressing three core challenges in complex video object segmentation—detection of small/low-resolution objects, robustness to severe occlusion, and dynamic scene modeling—this paper proposes the MASSeg framework and the MOSE+ enhanced dataset. Methodologically: (1) we introduce MOSE+, the first benchmark explicitly designed for occlusion-aware and motion-aware video segmentation; (2) we propose inter-frame consistency-aware data augmentation to explicitly model both consistent and inconsistent motion patterns; and (3) we devise an adaptive mask scaling inference mechanism that dynamically aligns mask resolution with multi-scale object sizes. The approach integrates multi-scale feature aggregation, cross-frame consistency regularization, and hybrid augmentation strategies. On the MOSE test set, MASSeg achieves J=0.8250, F=0.9007, and J&F=0.8628, securing second place in the MOSE track of the CVPR 2025 PVUW Challenge.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Mixture of Experts (MoE)
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Complex video object segmentation continues to face significant challenges in small object recognition, occlusion handling, and dynamic scene modeling. This report presents our solution, which ranked second in the MOSE track of CVPR 2025 PVUW Challenge. Based on an existing segmentation framework, we propose an improved model named MASSeg for complex video object segmentation, and construct an enhanced dataset, MOSE+, which includes typical scenarios with occlusions, cluttered backgrounds, and small target instances. During training, we incorporate a combination of inter-frame consistent and inconsistent data augmentation strategies to improve robustness and generalization. During inference, we design a mask output scaling strategy to better adapt to varying object sizes and occlusion levels. As a result, MASSeg achieves a J score of 0.8250, F score of 0.9007, and a J&F score of 0.8628 on the MOSE test set.

Problem

• Improving small object recognition in video segmentation
• Enhancing occlusion handling in dynamic scenes
• Developing robust models for cluttered backgrounds

Innovation

• Enhanced dataset MOSE+ with occlusion scenarios
• Inter-frame consistent and inconsistent augmentation
• Mask output scaling for varying objects