Convolutions Die Hard: Open-Vocabulary Segmentation with Single Frozen Convolutional CLIP

Benchmark Model Rank Results
open-vocabulary-semantic-segmentation-on-cityscapesFC-CLIP#1mIoU: 56.2
open-vocabulary-semantic-segmentation-on-pascalvoc-20bFC-CLIP#3mIoU: 81.8
open-vocabulary-panoptic-segmentation-on-ade20kFC-CLIP#6PQ: 26.8
open-vocabulary-semantic-segmentation-on-pascalvoc-20FC-CLIP#8mIoU: 95.4
open-vocabulary-semantic-segmentation-on-ade20k-847FC-CLIP#10mIoU: 14.8
open-vocabulary-semantic-segmentation-on-pascal-context-459FC-CLIP#11mIoU: 18.2
open-vocabulary-semantic-segmentation-on-ade20k-150FC-CLIP#13mIoU: 34.1
open-vocabulary-semantic-segmentation-on-pascal-context-59FC-CLIP#14mIoU: 58.4