Grounding Everything: Emerging Localization Properties in Vision-Language Transformers

Open paper

Benchmark	Model	Rank	Results
zero-shot-segmentation-on-ade20k-training-free	GEM (MetaCLIP)	#2	mIoU: 17.1
zero-shot-segmentation-on-ade20k-training-free	GEM (CLIP)	#3	mIoU: 15.7