Semantic Segmentation on Cityscapes (mIoU, mF1, mPre, mRec)
44.74mIoUMoE-RAM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MoE-RAMbackbone=Frozen ViT2025.12 | 44.74 | 55.99 | 68.3 | 53.47 | |
| SegNet2025.12 | 43.13 | 53.87 | — | — | |
| SoftMoEbackbone=Frozen ViT2025.12 | 42.08 | 54.45 | 66.93 | 49.79 | |
| NonLinearMoEbackbone=Frozen ViT2025.12 | 41.74 | 54.07 | 65.9 | 50.15 | |
| LinearMoEbackbone=Frozen ViT2025.12 | 39.8 | 51.53 | 66.07 | 47.62 | |
| SegFormer2025.12 | 39.37 | 46.23 | — | — | |
| HRDA2025.12 | 38.89 | 49.38 | 64.3 | 45.8 | |
| BiSecNetV22025.12 | 33.63 | 43.32 | — | — | |
| ViT+ASSPbackbone=Frozen ViT2025.12 | 26.31 | 30.37 | — | — | |
| AttaNet2025.12 | 22.96 | 27.28 | 26.11 | 30.87 | |
| TopFormer2025.12 | 22.15 | 26.7 | 25.48 | 30.24 | |
| SeaFormer2025.12 | 20.51 | 24.02 | 23.36 | 26.85 |