Semantic Segmentation on Cityscapes (val) (mIoU, GFLOPs, Params)
138GFLOPsHRViT-b2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HRViT-b2Resolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 138 | 82.8 | — | — | |
| RSSeg-CSWin-TResolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 170 | 83.41 | 84.58 | — | |
| RSSeg-CSWin-SResolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 262 | 83.64 | 84.56 | — | |
| SegFormer-B2Resolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 290 | 81 | 82.2 | — | |
| HRViT-b3Resolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 307 | 83.16 | — | — | |
| RSSeg-ViT-BResolution=768x768, Backbone Type=Plain ViT, Pre-train Weights=ViT-AugReg [26]2022.12 | 338 | 81.2 | 83.26 | — | |
| Segmenter-BResolution=768x768, Backbone Type=Plain ViT2022.12 | 350 | — | 80.6 | — | |
| HRNetResolution=1024x1024, Backbone Type=CNN2022.12 | 375 | 80.7 | 81.9 | — | |
| SegFormer-B5Resolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 562 | 82.4 | 84 | — | |
| OCRNetResolution=1024x1024, Backbone Type=CNN2022.12 | 649 | 81.4 | 82.7 | — | |
| EncNetResolution=1024x1024, Backbone Type=CNN2022.12 | 874 | 78.6 | 79.5 | — | |
| lawin-LResolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 899 | — | 84.4 | — | |
| Semask-L-Mask2Resolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 900 | 83.97 | 84.98 | — | |
| DeepLabV3+Resolution=1024x1024, Backbone Type=CNN2022.12 | 1,016 | 81 | 82.2 | — | |
| PSPNetResolution=1024x1024, Backbone Type=CNN2022.12 | 1,025 | 79.8 | 81.1 | — | |
| RSSeg-ViT-LResolution=768x768, Backbone Type=Plain ViT, Pre-train Weights=ViT-AugReg [26]2022.12 | 1,039 | 81.3 | 83.45 | — | |
| Segmenter-LResolution=768x768, Backbone Type=Plain ViT, Pre-train Weights=ViT-AugReg [26]2022.12 | 1,045 | 79.1 | 81.3 | — | |
| StructToken-PWE-LResolution=768x768, Backbone Type=Plain ViT, Pre-train Weights=ViT-AugReg [26]2022.12 | 1,050 | 80.05 | 82.07 | — | |
| SETR-PUP-LResolution=768x768, Backbone Type=Plain ViT2022.12 | 1,082 | 79.34 | 82.15 | — | |
| FCNResolution=1024x1024, Backbone Type=CNN2022.12 | 1,102 | 75.5 | 76.6 | — | |
| DANetResolution=1024x1024, Backbone Type=CNN2022.12 | 1,108 | 80.5 | 82 | — | |
| CCNetResolution=1024x1024, Backbone Type=CNN2022.12 | 1,113 | 79.5 | 80.7 | — | |
| HRFormer-B + OCRResolution=1024x1024, Backbone Type=Pyramid ViT2022.12 | 1,120 | 81.9 | 82.6 | — |