Semantic Segmentation on Cityscapes (mIoU Drop Analysis)
90.35mIoUClean Image
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Clean ImageModel=PIDNet-L2026.03 | 90.35 | — | — | — | |
| Random PatchModel=PIDNet-L2026.03 | 89.96 | — | — | — | |
| Baseline PatchModel=PIDNet-L2026.03 | 89.84 | — | — | — | |
| Clean ImageModel=PIDNet-S2026.03 | 86.95 | — | — | — | |
| Clean ImageModel=PIDNet-M2026.03 | 86.81 | — | — | — | |
| Random PatchModel=PIDNet-S2026.03 | 86.51 | — | — | — | |
| Random PatchModel=PIDNet-M2026.03 | 86.18 | — | — | — | |
| Baseline PatchModel=PIDNet-M2026.03 | 86.15 | — | — | — | |
| Baseline PatchModel=PIDNet-S2026.03 | 77.91 | — | — | — | |
| OmniPatchModel=PIDNet-L2026.03 | 75.3 | 16.65 | 16.3 | 16.18 | |
| Clean ImageModel=SegFormer, Input Scaling=Downscaled2026.03 | 74.34 | — | — | — | |
| Random PatchModel=SegFormer, Input Scaling=Downscaled2026.03 | 74.31 | — | — | — | |
| OmniPatchModel=PIDNet-M2026.03 | 73.93 | 14.84 | 14.22 | 14.18 | |
| OmniPatchModel=PIDNet-S2026.03 | 72.99 | 16.05 | 15.62 | 6.31 | |
| Clean ImageModel=BiSeNetv12026.03 | 71.49 | — | — | — | |
| Baseline PatchModel=BiSeNetv12026.03 | 71.2 | — | — | — | |
| Random PatchModel=BiSeNetv12026.03 | 70.57 | — | — | — | |
| Clean ImageModel=BiSeNetv22026.03 | 69.07 | — | — | — | |
| Baseline PatchModel=BiSeNetv22026.03 | 68.48 | — | — | — | |
| Random PatchModel=BiSeNetv22026.03 | 68.45 | — | — | — | |
| OmniPatchModel=SegFormer, Input Scaling=Downscaled2026.03 | 67.77 | 8.83 | 8.79 | — | |
| OmniPatchModel=BiSeNetv12026.03 | 64.1 | 10.33 | 9.17 | 9.97 | |
| OmniPatchModel=BiSeNetv22026.03 | 60.36 | 12.61 | 11.81 | 11.85 |