Semantic Segmentation on VSPW Dataset through Masked Video Consistency

TL;DR

This paper introduces masked video consistency (MVC) to improve spatiotemporal modeling in video semantic segmentation, achieving high accuracy on the VSPW dataset by enforcing prediction consistency on masked frames.

Contribution

We propose masked video consistency (MVC) to enhance spatiotemporal learning in video segmentation models, addressing limitations of previous models on the VSPW dataset.

Findings

Achieved 67.27% mIoU on VSPW dataset.

Ranked 2nd in PVUW2024 challenge VSS track.

Demonstrated effectiveness of MVC with test-time augmentation and multimodal post-processing.

Abstract

Pixel-level Video Understanding requires effectively integrating three-dimensional data in both spatial and temporal dimensions to learn accurate and stable semantic information from continuous frames. However, existing advanced models on the VSPW dataset have not fully modeled spatiotemporal relationships. In this paper, we present our solution for the PVUW competition, where we introduce masked video consistency (MVC) based on existing models. MVC enforces the consistency between predictions of masked frames where random patches are withheld. The model needs to learn the segmentation results of the masked parts through the context of images and the relationship between preceding and succeeding frames of the video. Additionally, we employed test-time augmentation, model aggeregation and a multimodal model-based post-processing method. Our approach achieves 67.27% mIoU performance on the VSPW dataset, ranking 2nd place in the PVUW2024 challenge VSS track.