Tracking Anything with Decoupled Video Segmentation

Benchmark Model Rank Results
open-world-video-segmentation-on-burst-valDEVA (Mask2Former)#1OWTA (all): 69.9OWTA (com): 75.2OWTA (unc): 41.5
open-world-video-segmentation-on-burst-valDEVA (EntitySeg)#2OWTA (all): 69.5OWTA (com): 73.3OWTA (unc): 50.5
referring-expression-segmentation-on-davisDEVA (ReferFormer)#3J&F 1st frame: 66.3
referring-expression-segmentation-on-refer-1DEVA (ReferFormer)#11J&F: 66.0
semi-supervised-video-object-segmentation-on-1DEVA#7J&F: 83.2Jaccard (Mean): 79.6F-measure (Mean): 86.8FPS: 25.3
semi-supervised-video-object-segmentation-on-21DEVA (with OVIS)#8J&F: 66.5J: 62.3F: 70.8FPS: 25.3
semi-supervised-video-object-segmentation-on-21DEVA (no OVIS)#11J&F: 60.0J: 55.8F: 64.3FPS: 25.3
unsupervised-video-object-segmentation-on-10DEVA (DIS)#2G: 88.9J: 87.6F: 90.2
unsupervised-video-object-segmentation-on-4DEVA (EntitySeg)#1J&F: 73.4Jaccard (Mean): 70.4F-measure (Mean): 76.4
unsupervised-video-object-segmentation-on-5DEVA (EntitySeg)#1J&F: 62.1
video-panoptic-segmentation-on-vipsegDEVA (Mask2Former - SwinB)#6VPQ: 55.0STQ: 52.2
visual-object-tracking-on-davis-2017DEVA#8J&F: 87.6Jaccard (Mean): 84.2F-measure (Mean): 91.0