Semantic Segmentation on ADE20K (mIoU, mAcc, aAcc)
57.1mIoUFull tuning
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Full tuningBackbone=DINOv22026.05 | 57.1 | 70.5 | 85.5 | |
| SIGMABackbone=DINOv22026.05 | 56.42 | 69.28 | 85.06 | |
| Full tuningBackbone=SigLIP22026.05 | 55.89 | 69.1 | 84.85 | |
| VFM-AdapterBackbone=DINOv22026.05 | 55.14 | 68.5 | 84.85 | |
| SIGMABackbone=SigLIP22026.05 | 55.12 | 68.23 | 84.72 | |
| LorandBackbone=DINOv22026.05 | 54.72 | 68.25 | 84.34 | |
| AdaptformerBackbone=DINOv22026.05 | 54.68 | 68.18 | 84.63 | |
| VFM-AdapterBackbone=SigLIP22026.05 | 54.36 | 67.85 | 84.45 | |
| LorandBackbone=SigLIP22026.05 | 53.84 | 67.42 | 84.53 | |
| AdaptformerBackbone=SigLIP22026.05 | 53.81 | 67.77 | 84.41 | |
| AdapterBackbone=SigLIP22026.05 | 53.77 | 67.07 | 84.35 | |
| AdapterBackbone=DINOv22026.05 | 53.76 | 67.92 | 84.46 | |
| Partial-1Backbone=SigLIP22026.05 | 53.56 | 67.26 | 84.12 | |
| Partial-1Backbone=DINOv22026.05 | 53.43 | 67.63 | 84.57 | |
| BitFitBackbone=SigLIP22026.05 | 52.71 | 67.93 | 84.21 | |
| LoRABackbone=DINOv22026.05 | 52.47 | 67.6 | 84.23 | |
| LoRABackbone=SigLIP22026.05 | 52.47 | 66.31 | 83.67 | |
| BitFitBackbone=DINOv22026.05 | 52.45 | 67.62 | 84.63 | |
| FixedBackbone=SigLIP22026.05 | 51.66 | 66.54 | 83.04 | |
| FixedBackbone=DINOv22026.05 | 51.6 | 67.8 | 83.7 | |
| Full tuningBackbone=SAM2026.05 | 49.54 | 61.45 | 82.35 | |
| SIGMABackbone=SAM2026.05 | 48.58 | 61.1 | 81.95 | |
| VFM-AdapterBackbone=SAM2026.05 | 47.36 | 60.25 | 81.6 | |
| Partial-1Backbone=SAM2026.05 | 47.18 | 60.1 | 81.55 | |
| LorandBackbone=SAM2026.05 | 47.05 | 60.05 | 81.52 | |
| DeiT-BVariant=Dense2026.07 | 47 | 57.5 | 82.6 | |
| AdaptformerBackbone=SAM2026.05 | 46.64 | 59.8 | 81.48 | |
| AdapterBackbone=SAM2026.05 | 46.52 | 59.75 | 81.45 | |
| HetDPTLayers pruned=8 layers, Pre-training=ImageNet-1k2026.07 | 46.4 | 56.5 | 82.2 | |
| LoRABackbone=SAM2026.05 | 46.36 | 59.65 | 81.4 | |
| BitFitBackbone=SAM2026.05 | 46.28 | 59.6 | 81.35 | |
| NOSELayers pruned=5 layers2026.07 | 46.2 | 56.5 | 82.2 | |
| FixedBackbone=SAM2026.05 | 45.72 | 59.32 | 81.21 | |
| NOSELayers pruned=6 layers2026.07 | 45.6 | 55.2 | 82 | |
| EViT2026.07 | 45.5 | 55.9 | 81.9 | |
| HetDPTLayers pruned=10 layers, Pre-training=ImageNet-1k2026.07 | 45.4 | 55.7 | 81.9 | |
| TPS2026.07 | 45.3 | 55.1 | 81.9 | |
| DINOArch=ViT-S, Pretrain=SSv2, Evaluation Framework=UperNet2026.06 | 10.71 | 14.64 | — | |
| iBOTArch=ViT-S, Pretrain=SSv2, Evaluation Framework=UperNet2026.06 | 10.6 | 14.53 | — | |
| TDVArch=ViT-S, Pretrain=SSv2, Evaluation Framework=UperNet2026.06 | 10.54 | 14.48 | — | |
| DINOArch=ViT-B, Pretrain=SSv2, Evaluation Framework=UperNet2026.06 | 10.48 | 11.14 | — | |
| iBOTArch=ViT-B, Pretrain=SSv2, Evaluation Framework=UperNet2026.06 | 9.94 | 13.65 | — | |
| TDVArch=ViT-B, Pretrain=SSv2, Evaluation Framework=UperNet2026.06 | 9.57 | 10.7 | — |