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

Benchmark Model Rank Results
open-vocabulary-panoptic-segmentation-on-ade20kFC-CLIP#5PQ: 26.8
open-vocabulary-semantic-segmentation-on-1FC-CLIP#12mIoU: 58.4
open-vocabulary-semantic-segmentation-on-2FC-CLIP#9mIoU: 34.1
open-vocabulary-semantic-segmentation-on-3FC-CLIP#8mIoU: 14.8
open-vocabulary-semantic-segmentation-on-5FC-CLIP#7mIoU: 95.4
open-vocabulary-semantic-segmentation-on-7FC-CLIP#9mIoU: 18.2
open-vocabulary-semantic-segmentation-on-9FC-CLIP#3mIoU: 81.8
open-vocabulary-semantic-segmentation-on-cityscapesFC-CLIP#1mIoU: 56.2