Semantic Segmentation on Cityscapes 8 classes (VKITTI2 context)
79.63mIoUMTL target
Evaluation Results
| Method | Links | |
|---|---|---|
| MTL targetArchitecture=Multi-Task Learning, Strategy=Oracle (Target Labels)2022.06 | 79.63 | |
| MTL targetBackbone=DeepLabV2, UDA setting=False, Learning protocol (STL/MTL)=MTL, Source/Target domain evaluation protocol=Target-supervised (Upper Bound)2025.09 | 79.63 | |
| STL targetArchitecture=Single-Task Learning, Strategy=Oracle (Target Labels)2022.06 | 77.1 | |
| STL targetBackbone=DeepLabV2, UDA setting=False, Learning protocol (STL/MTL)=STL, Source/Target domain evaluation protocol=Target-supervised (Upper Bound)2025.09 | 77.1 | |
| VTAGMLBackbone=Swin, UDA setting=True, Learning protocol (STL/MTL)=MTL2025.09 | 70.93 | |
| FAMDABackbone=SegFormer, UDA setting=True, Learning protocol (STL/MTL)=MTL2025.09 | 68.4 | |
| MulTBackbone=Swin, UDA setting=True, Learning protocol (STL/MTL)=MTL2025.09 | 66.12 | |
| FAMDABackbone=DeepLabV2, UDA setting=True, Learning protocol (STL/MTL)=MTL2025.09 | 65.86 | |
| Swin-UDABackbone=Swin, UDA setting=True, Learning protocol (STL/MTL)=STL2025.09 | 63.88 | |
| mTEB (Ours)Architecture=Multi-Task Learning with mTEB, Strategy=Unsupervised Domain Adaptation2022.06 | 63.76 | |
| XTAMBackbone=DeepLabV2, UDA setting=True, Learning protocol (STL/MTL)=MTL2025.09 | 63.76 | |
| STL-UDAArchitecture=Single-Task Learning, Strategy=Unsupervised Domain Adaptation2022.06 | 61.6 | |
| STL-UDABackbone=DeepLabV2, UDA setting=True, Learning protocol (STL/MTL)=STL2025.09 | 61.6 | |
| STL sourceArchitecture=Single-Task Learning, Strategy=Source only2022.06 | 58.77 | |
| STL sourceBackbone=DeepLabV2, UDA setting=False, Learning protocol (STL/MTL)=STL, Source/Target domain evaluation protocol=Source-only (No adaptation)2025.09 | 58.77 | |
| MTL-UDAArchitecture=Multi-Task Learning, Strategy=Unsupervised Domain Adaptation2022.06 | 57.26 | |
| MTL-UDABackbone=DeepLabV2, UDA setting=True, Learning protocol (STL/MTL)=MTL2025.09 | 57.26 | |
| MTL sourceArchitecture=Multi-Task Learning, Strategy=Source only2022.06 | 49.5 | |
| MTL sourceBackbone=DeepLabV2, UDA setting=False, Learning protocol (STL/MTL)=MTL, Source/Target domain evaluation protocol=Source-only (No adaptation)2025.09 | 49.5 |