VISA: Reasoning Video Object Segmentation via Large Language Models

Benchmark Model Rank Results
referring-video-object-segmentation-on-refvos-davisVISA Chat-UniVi-13B#6J&F: 70.4
referring-video-object-segmentation-on-ref-davis17VISA Chat-UniVi-13B (extra data)#9J&F: 70.4
referring-video-object-segmentation-on-revosVISA (Chat-UniVi-13B)#19J&F: 50.9J: 48.8F: 52.9R: 14.5
referring-video-object-segmentation-on-revosVISA (Chat-UniVi-7B)#23J&F: 46.9J: 44.9F: 49.0R: 15.5
referring-video-object-segmentation-on-mevisVISA Chat-UniVi-13B (extra data)#26J&F: 44.5
referring-expression-segmentation-on-refer-youtube-vos-2021-public-validationVISA Chat-UniVi-13B (extra data)#32J&F: 63.0