VideoSEG-O3: A Multi-turn Reinforcement Learning Framework for Reasoning Video Object Segmentation

Benchmark Model Rank Results
referring-video-object-segmentation-on-long-rvosVideoSEG-O3-4B#1J&F: 57.4
referring-video-object-segmentation-on-long-rvosVideoSEG-O3-2B#2J&F: 54.8
referring-expression-segmentation-on-refer-youtube-vos-2021-public-validationVideoSEG-O3-4B#2J&F: 74.1
referring-expression-segmentation-on-refer-youtube-vos-2021-public-validationVideoSEG-O3-2B#11J&F: 70.5
referring-video-object-segmentation-on-ref-davis17VideoSEG-O3-2B#3J&F: 80.0
referring-video-object-segmentation-on-ref-davis17VideoSEG-O3-4B#6J&F: 79.4
referring-video-object-segmentation-on-revosVideoSEG-O3-4B#4J&F: 67.7J: 65.0F: 70.4
referring-video-object-segmentation-on-revosVideoSEG-O3-2B#6J&F: 64.8J: 62.0F: 67.5