CorrCLIP: Reconstructing Correlations in CLIP with Off-the-Shelf Foundation Models for Open-Vocabulary Semantic Segmentation

Benchmark Model Rank Results
unsupervised-semantic-segmentation-with-language-image-pre-training-on-cityscapes-valCorrCLIP#1mIoU: 51.1
unsupervised-semantic-segmentation-with-language-image-pre-training-on-coco-objectCorrCLIP#1mIoU: 49.4
unsupervised-semantic-segmentation-with-language-image-pre-training-on-pascal-context-59CorrCLIP#1mIoU: 50.8
unsupervised-semantic-segmentation-with-language-image-pre-training-on-pascal-context-60CorrCLIP#1mIoU: 44.9
unsupervised-semantic-segmentation-with-language-image-pre-training-on-ade20kCorrCLIP#2Mean IoU (val): 30.7
unsupervised-semantic-segmentation-with-language-image-pre-training-on-pascal-vocCorrCLIP#2mIoU: 76.7
unsupervised-semantic-segmentation-with-language-image-pre-training-on-coco-stuff-171CorrCLIP#3mIoU: 34.0
unsupervised-semantic-segmentation-with-language-image-pre-training-on-pascalvoc-20CorrCLIP#3mIoU: 91.8