Panoptic Segmentation on COCO panoptic 133 categories (val)
52.7PQMaskFormer
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MaskFormerbackbone=Swin-L*, Pre-training=ImageNet-22K, #params=212M, FLOPS=792G, fps=5.22021.07 | 52.7 | 58.5 | 44 | 81.8 | 63.5 | |
| MaskFormerbackbone=Swin-B*, Pre-training=ImageNet-22K, #params=102M, FLOPS=411G, fps=8.42021.07 | 51.8 | 56.9 | 44.1 | 81.4 | 62.6 | |
| Max-DeepLabbackbone=Max-L, #params=451M, FLOPS=3692G2021.07 | 51.1 | 57 | 42.2 | — | — | |
| MaskFormerbackbone=Swin-B, #params=102M, FLOPS=411G, fps=8.42021.07 | 51.1 | 56.3 | 43.2 | 81.4 | 61.8 | |
| MaskFormerbackbone=Swin-S, #params=63M, FLOPS=259G, fps=12.42021.07 | 49.7 | 54.4 | 42.6 | 80.9 | 60.4 | |
| Max-DeepLabbackbone=Max-S, #params=62M, FLOPS=324G, fps=7.62021.07 | 48.4 | 53 | 41.5 | — | — | |
| MaskFormerbackbone=Swin-T, #params=42M, FLOPS=179G, fps=172021.07 | 47.7 | 51.7 | 41.7 | 80.4 | 58.3 | |
| MaskFormerbackbone=R101 + 6 Enc, #params=64M, FLOPS=248G, fps=142021.07 | 47.6 | 52.5 | 40.3 | 80.7 | 58 | |
| MaskFormerbackbone=R50 + 6 Enc, #params=45M, FLOPS=181G, fps=17.62021.07 | 46.5 | 51 | 39.8 | 80.4 | 56.8 | |
| MaskFormer (DETR post-processing)backbone=R50 + 6 Enc, post-processing=DETR-style2021.07 | 45.6 | 50 | 39 | 80.2 | 55.8 | |
| DETRbackbone=R101 + 6 Enc2021.07 | 45.1 | 50.5 | 37 | 79.9 | 55.5 | |
| DETRbackbone=R50 + 6 Enc2021.07 | 43.4 | 48.2 | 36.3 | 79.3 | 53.8 |