Semantic Segmentation on Cityscapes (mIoU, GFLOPs, FPS)
75.2mIoUSegViT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SegViTBackbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 75.2 | 670 | 24 | |
| +DToPBackbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 75 | 430 | 16 | |
| +CTS & DToPBackbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 75 | 296 | 21 | |
| +CTSBackbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 74.9 | 403 | 46 | |
| SegViTBackbone=ViT-Base, Input Resolution=768 x 7682025.09 | 73.7 | 301 | 65 | |
| +DToPBackbone=ViT-Base, Input Resolution=768 x 7682025.09 | 73.5 | 198 | 22 | |
| +CTSBackbone=ViT-Base, Input Resolution=768 x 7682025.09 | 72.9 | 190 | 45 | |
| +dCTS τ-6899Backbone=ViT-Base, Input Resolution=768 x 7682025.09 | 72.8 | 182 | 68 | |
| +CTS & DToPBackbone=ViT-Base, Input Resolution=768 x 7682025.09 | 72.7 | 135 | 22 | |
| +dCTS τ-6899Backbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 72.7 | 247 | 62 | |
| +STEP@[6,8]τ-6899Backbone=ViT-Base, Input Resolution=768 x 7682025.09 | 72.6 | 131 | 21 | |
| +STEP@[6,8]τ-6899Backbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 72 | 183 | 25 | |
| +STEP@[8]τ-6899Backbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 72 | 199 | 35 | |
| +STEP@[8]τ-4999Backbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 71.1 | 163 | 36 | |
| +STEP@[6,8]τ-4999Backbone=ViT-Base, Input Resolution=1024 x 10242025.09 | 71 | 149 | 25 | |
| +STEP@[8]τ-6899Backbone=ViT-Base, Input Resolution=768 x 7682025.09 | 69.9 | 149 | 28 | |
| +STEP@[8]τ-4999Backbone=ViT-Base, Input Resolution=768 x 7682025.09 | 69.9 | 105 | 22 | |
| +STEP@[6,8]τ-4999Backbone=ViT-Base, Input Resolution=768 x 7682025.09 | 69.8 | 95 | 22 | |
| SegViT + DToPBackbone=ViT-Base, Input Resolution=512 x 512, Configuration=Default from original paper2025.09 | 68.2 | 82 | 24 | |
| SegViTBackbone=ViT-Base, Input Resolution=512 x 5122025.09 | 67.7 | 110 | 70 | |
| SegViT + CTSBackbone=ViT-Base, Input Resolution=512 x 512, Configuration=Default from original paper2025.09 | 67.6 | 73 | 56 | |
| SegViT + CTS & DToPBackbone=ViT-Base, Input Resolution=512 x 512, Configuration=Default from original paper2025.09 | 67.5 | 59.5 | 27 | |
| SegViT + dCTS τ-6899Backbone=ViT-Base, Input Resolution=512 x 5122025.09 | 67.5 | 77 | 66 | |
| SegViT + STEP@[8]τ-6899Backbone=ViT-Base, Input Resolution=512 x 5122025.09 | 67.4 | 66 | 32 | |
| SegViT + STEP@[6,8]τ-6899Backbone=ViT-Base, Input Resolution=512 x 5122025.09 | 67.2 | 61 | 22 | |
| SegViT + STEP@[6,8]τ-4999Backbone=ViT-Base, Input Resolution=512 x 5122025.09 | 64.3 | 44 | 26 | |
| SegViT + STEP@[8]τ-4999Backbone=ViT-Base, Input Resolution=512 x 5122025.09 | 64.2 | 44 | 32 |