Semantic Segmentation on VSPW Dataset through Masked Video Consistency
This work addresses pixel-level video understanding for researchers, but it is incremental as it builds on existing models with added techniques like MVC and post-processing.
The paper tackles the problem of incomplete spatiotemporal modeling in semantic segmentation on the VSPW dataset by introducing masked video consistency (MVC) to enforce prediction consistency in masked frames, achieving 67.27% mIoU and ranking 2nd in the PVUW2024 challenge.
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.