UMG-CLIP: A Unified Multi-Granularity Vision Generalist for Open-World Understanding

Benchmark Model Rank Results
open-vocabulary-panoptic-segmentation-on-ade20kUMG-CLIP-E/14#1PQ: 31.6
open-vocabulary-panoptic-segmentation-on-ade20kUMG-CLIP-L/14#3PQ: 29.1
open-vocabulary-semantic-segmentation-on-1UMG-CLIP-L/14#5mIoU: 61.0
open-vocabulary-semantic-segmentation-on-2UMG-CLIP-E/14#1mIoU: 38.2
open-vocabulary-semantic-segmentation-on-2UMG-CLIP-L/14#5mIoU: 36.1
open-vocabulary-semantic-segmentation-on-3UMG-CLIP-E/14#1mIoU: 17.3
open-vocabulary-semantic-segmentation-on-3UMG-CLIP-L/14#5mIoU: 15.4
open-vocabulary-semantic-segmentation-on-5UMG-CLIP-L/14#1mIoU: 97.9
open-vocabulary-semantic-segmentation-on-7UMG-CLIP-E/14#1mIoU: 25.2
open-vocabulary-semantic-segmentation-on-7UMG-CLIP-L/14#4mIoU: 23.2
open-vocabulary-semantic-segmentation-on-9UMG-CLIP-E/14#1mIoU: 85.4
panoptic-segmentation-on-coco-minivalUMG-CLIP-E/14#4PQ: 59.5AP: 50.7mIoU: 69.7
panoptic-segmentation-on-coco-minivalUMG-CLIP-L/14#7PQ: 58.9AP: 49.7mIoU: 68.9