Open-Vocabulary Panoptic Segmentation with Text-to-Image Diffusion Models

Benchmark Model Rank Results
zero-shot-segmentation-on-segmentation-in-the-wildodise#6Mean AP: 38.7
open-vocabulary-panoptic-segmentation-on-ade20kODISE(Caption)#8PQ: 23.4
open-vocabulary-panoptic-segmentation-on-ade20kODISE (Label)#9PQ: 22.6
open-vocabulary-semantic-segmentation-on-pascal-context-459ODISE#15mIoU: 14.5
open-vocabulary-semantic-segmentation-on-pascal-context-59ODISE#15mIoU: 57.3
open-vocabulary-semantic-segmentation-on-pascalvoc-20ODISE#15mIoU: 84.6
open-vocabulary-semantic-segmentation-on-ade20k-150ODISE#18mIoU: 29.9
open-vocabulary-semantic-segmentation-on-ade20k-847ODISE#18mIoU: 11.1