Semantic Segmentation on COCO (mIoU, Inference Time, Speedup)
66.03mIoUMhSA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MhSAViT=Large, Attention=MhSA, |S|=-2026.05 | 66.03 | — | 161.4 | — | |
| DW[all]ViT=Large, Attention=DW[all], |S|=12/242026.05 | 65.95 | -0.08 | 133.6 | 17.21 | |
| Conv[all]ViT=Large, Attention=Conv[all], |S|=12/242026.05 | 65.84 | -0.19 | 173.5 | -7.49 | |
| SPViT-styleViT=Large, Attention=SPViT-style, |S|=10/242026.05 | 65.05 | -0.98 | 133.3 | 17.4 | |
| Ens + DWViT=Large, Attention=Ens + DW, |S|=10/242026.05 | 64.81 | -1.61 | 131.1 | 18.79 | |
| DW[all]ViT=Large, Attention=DW[all], |S|=16/242026.05 | 64.64 | -1.39 | 126.1 | 21.85 | |
| Ens + DWViT=Large, Attention=Ens + DW, |S|=12/242026.05 | 63.92 | -2.11 | 124.5 | 22.87 | |
| MhSAViT=Base, Attention=MhSA, |S|=-2026.05 | 63.37 | — | 49.9 | — | |
| DW[all]ViT=Base, Attention=DW[all], |S|=6/122026.05 | 62.22 | -1.16 | 40.8 | 18.1 | |
| Ens + DWViT=Base, Attention=Ens + DW, |S|=6/122026.05 | 60.55 | -2.83 | 38.1 | 23.52 |