Recent advancements in video semantic segmentation have made substantial
progress by exploiting temporal correlations. Nevertheless, persistent
challenges, including redundant computation and the reliability of the feature
propagation process, underscore the need for further innovation