HyperSeg: Towards Universal Visual Segmentation with Large Language Model

Benchmark Model Rank Results
panoptic-segmentation-on-coco-minivalHyperSeg (Swin-B)#1PQ: 61.2
referring-expression-segmentation-on-refcoco-plus-testaHyperSeg#1Overall IoU: 83.5
referring-expression-segmentation-on-refcoco-testbHyperSeg#1Overall IoU: 83.4
semantic-segmentation-on-cocoHyperSeg#1mIoU: 77.2
open-vocabulary-semantic-segmentation-on-pascal-context-59HyperSeg#2mIoU: 64.6
referring-expression-segmentation-on-davis-2017-valHyperSeg#2J&F 1st frame: 71.2
referring-expression-segmentation-on-refcoco-testaHyperSeg#2Overall IoU: 85.7
referring-expression-segmentation-on-refcoco-plus-test-bHyperSeg#3Overall IoU: 75.2
referring-expression-segmentation-on-refcoco-valHyperSeg#3Overall IoU: 84.8
referring-expression-segmentation-on-refcocog-valHyperSeg#3Overall IoU: 79.4
referring-expression-segmentation-on-refcoco-plus-valHyperSeg#4Overall IoU: 79.0
referring-video-object-segmentation-on-refer-youtube-vosHyperSeg#4J&F: 68.5
referring-expression-segmentation-on-refcocog-testHyperSeg#5Overall IoU: 78.9
open-vocabulary-semantic-segmentation-on-pascalvoc-20HyperSeg#12mIoU: 92.1