Panoptic Segmentation on COCO 2017 (minival)
57.9PQMask2Former (FocalNet-L†, 200 queries)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Mask2Former (FocalNet-L†, 200 queries)Backbone=FocalNet-L, Segmentation Method=Mask2Former, Queries=200, Parameters=226M, Pre-training=ImageNet-22K, Scale=single-scale2022.03 | 57.9 | 48.4 | 67.3 | |
| Mask2Former (Swin-L†, 200 queries)Backbone=Swin-L, Segmentation Method=Mask2Former, Queries=200, Parameters=216M, Pre-training=ImageNet-22K, Scale=single-scale2022.03 | 57.8 | 48.6 | 67.4 | |
| Panoptic SegFormer (Swin-L†)Backbone=Swin-L, Segmentation Method=Panoptic SegFormer, Pre-training=ImageNet-22K, Scale=single-scale2022.03 | 55.8 | — | — | |
| Mask2Former (Swin-B)Backbone=Swin-B, Segmentation Method=Mask2Former, Parameters=107M, Scale=single-scale2022.03 | 55.1 | 45.2 | 65.1 | |
| K-Net (Swin-L)Backbone=Swin-L, Segmentation Method=K-Net, Scale=single-scale2022.03 | 54.3 | — | — | |
| Panoptic SegFormer (PVTv2-B5)Backbone=PVTv2-B5, Segmentation Method=Panoptic SegFormer, Parameters=101M, Scale=single-scale2022.03 | 54.1 | — | — | |
| Mask2Former (Swin-T)Backbone=Swin-T, Segmentation Method=Mask2Former, Parameters=47M, Scale=single-scale2022.03 | 53.2 | 43.3 | 63.2 | |
| MaskFormer (Swin-L†)Backbone=Swin-L, Segmentation Method=MaskFormer, Parameters=212M, Pre-training=ImageNet-22K, Scale=single-scale2022.03 | 52.7 | 40.1 | 64.8 | |
| Mask2Former (ResNet-50)Backbone=ResNet-50, Segmentation Method=Mask2Former, Parameters=44M, Scale=single-scale2022.03 | 51.9 | 41.7 | 62.4 | |
| MaskFormer (Swin-B)Backbone=Swin-B, Segmentation Method=MaskFormer, Parameters=102M, Scale=single-scale2022.03 | 51.1 | 37.8 | 62.6 | |
| Panoptic SegFormer (ResNet-50)Backbone=ResNet-50, Segmentation Method=Panoptic SegFormer, Parameters=47M, Scale=single-scale2022.03 | 50 | — | — | |
| MaskFormer (Swin-T)Backbone=Swin-T, Segmentation Method=MaskFormer, Parameters=42M, Scale=single-scale2022.03 | 47.7 | 33.6 | 60.4 | |
| K-Net (ResNet-50)Backbone=ResNet-50, Segmentation Method=K-Net, Scale=single-scale2022.03 | 47.1 | — | — | |
| DETR (ResNet-50)Backbone=ResNet-50, Segmentation Method=DETR, Scale=single-scale2022.03 | 43.4 | — | — |