Per-Pixel Classification is Not All You Need for Semantic Segmentation

Open paper
Benchmark Model Rank Results
panoptic-segmentation-on-ade20k-valMaskFormer (R101 + 6 Enc)#24PQ: 35.7
panoptic-segmentation-on-coco-minivalMaskFormer (single-scale)#21PQ: 52.7PQst: 44.0PQth: 58.5RQ: 63.5SQ: 81.8
panoptic-segmentation-on-coco-test-devMaskFormer (Swin-L)#9PQ: 53.3PQst: 44.5PQth: 59.1
semantic-segmentation-on-ade20kMaskFormer(Swin-B)#70Validation mIoU: 53.8
semantic-segmentation-on-ade20kMaskFormer(ResNet-101)#148Validation mIoU: 48.1
semantic-segmentation-on-ade20k-valMaskFormer (Swin-L, ImageNet-22k pretrain)#27mIoU: 55.6
semantic-segmentation-on-mapillary-valMaskFormer (ResNet-50)#4mIoU: 55.4

Rank counts only results with a code link.