Deforming Videos to Masks: Flow Matching for Referring Video Segmentation

Benchmark Model Rank Results
referring-video-object-segmentation-on-mevisFlowRVS (Wan 2.1 T2V 1.3B DiT, MeViS-only) (extra data)#12J&F: 51.1
referring-expression-segmentation-on-refer-youtube-vos-2021-public-validationFlowRVS (Wan 2.1 T2V 1.3B DiT, RefCOCO pretrained) (extra data)#14J&F: 69.6
referring-video-object-segmentation-on-ref-davis17FlowRVS (Wan 2.1 T2V 1.3B DiT, zero-shot) (extra data)#15J&F: 73.3