Surface Normal Estimation on NYU v2
-22.1Mean Angular ErrorMTLMoE
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MTLMoEBackbone=Swin-Tiny, FLOPs (G)=77.16, Params (M)=115.902026.01 | -22.1 | 34.63 | — | — | — | — | — | |
| MeteoraBackbone=Swin-Tiny, FLOPs (G)=60.44, Params (M)=98.572026.01 | -17.8 | 32.92 | — | — | — | — | — | |
| PCGRADBackbone=Swin-Tiny, FLOPs (G)=21.50, Params (M)=29.502026.01 | -8.27 | 31.8 | — | — | — | — | — | |
| CAGRADBackbone=Swin-Tiny, FLOPs (G)=21.50, Params (M)=29.502026.01 | -7.4 | 33.68 | — | — | — | — | — | |
| Nash-MTLBackbone=Swin-Tiny, FLOPs (G)=21.50, Params (M)=29.502026.01 | -6.4 | 32.4 | — | — | — | — | — | |
| MTLoRABackbone=Swin-Tiny, FLOPs (G)=20.80, Params (M)=34.262026.01 | -3.2 | 30.27 | — | — | — | — | — | |
| Single TaskBackbone=Swin-Tiny, FLOPs (G)=73.70, Params (M)=112.602026.01 | 0 | 29.84 | — | — | — | — | — | |
| RobuMTLBackbone=Swin-Tiny, FLOPs (G)=21.50, Params (M)=30.932026.01 | 1.3 | 28.9 | — | — | — | — | — | |
| PAD-NetBackbone=HRNet-18, FLOPs (G)=310.70, Params (M)=19.522026.01 | 9.2 | 21.9 | — | — | — | — | — | |
| RobuMTL+Backbone=Swin-Tiny, FLOPs (G)=21.50, Params (M)=30.932026.01 | 9.7 | 27.52 | — | — | — | — | — | |
| Metric3Dv2Generative Model=×, MTL=✓, Inference Steps=12026.06 | 13.2 | — | 66.2 | — | — | — | — | |
| DINOv2 ViT-B/14Backbone=ViT-B/14, Evaluation Head=DPT head2026.02 | 14.1 | 21.7 | 63.4 | 80.8 | 86.5 | — | 134.9 | |
| Omnivorous ViT-B/14Backbone=ViT-B/14, Evaluation Head=DPT head2026.02 | 14.1 | 21.6 | 63.5 | 80.8 | 86.5 | — | 134.1 | |
| SenseNova-Vision2026.07 | 14.4 | — | — | — | — | — | — | |
| SenseNova-VisionMethod category=Generation-based2026.07 | 14.4 | — | 62.7 | — | — | — | — | |
| MoGe-2Method category=Geometry-specialized2026.07 | 14.7 | — | 62.3 | — | — | — | — | |
| ViGeo2026.05 | 15.11 | — | 61.3 | — | — | 8.51 | — | |
| OrchidGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 15.2 | — | 60.6 | — | — | — | — | |
| PRISM-XInference Mode=Zero-shot2025.12 | 15.7 | — | 57 | — | — | — | — | |
| NormalCrafter2026.05 | 15.87 | — | 60.41 | — | — | 8.7 | — | |
| Marigold-Normals v1.1Inference Mode=Zero-shot2025.12 | 16.1 | — | 60.5 | — | — | — | — | |
| PRISMInference Mode=Zero-shot2025.12 | 16.1 | — | 56.8 | — | — | — | — | |
| Marigold v1.1Training Data=77K, Zero-shot=true2025.09 | 16.1 | — | 60.5 | — | — | — | — | |
| Lotus-DTraining Data=59K, Zero-shot=true2025.09 | 16.2 | — | 59.8 | — | — | — | — | |
| FE2ETraining Data=71K, Zero-shot=true2025.09 | 16.2 | — | 59.6 | — | — | — | — | |
| EESNU2026.05 | 16.2 | — | — | 77.2 | 83.5 | 8.5 | — | |
| FE2EGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 16.2 | — | 59.6 | — | — | — | — | |
| FE2EMethod category=Generation-based2026.07 | 16.2 | — | 59.6 | — | — | — | — | |
| DSINEInference Mode=Zero-shot2025.12 | 16.4 | — | 59.6 | — | — | — | — | |
| DSINETraining Data=160K, Zero-shot=true2025.09 | 16.4 | — | 59.6 | — | — | — | — | |
| DSine2026.05 | 16.4 | — | — | 77.7 | 83.5 | 8.4 | — | |
| DSINETraining Data=160K, Zero-shot=true2026.06 | 16.4 | — | 59.6 | — | 83.5 | — | — | |
| UNIGPTraining Data=59K, Zero-shot=true2026.06 | 16.4 | — | 59.2 | — | 83.4 | — | — | |
| Marigoldv1.1Generative Model=✓, MTL=×, Inference Steps=12026.06 | 16.4 | — | 58.9 | — | — | — | — | |
| DSINEMethod category=Geometry-specialized2026.07 | 16.4 | — | 59.6 | — | — | — | — | |
| Lotus-GTraining Data=59K, Zero-shot=true2025.09 | 16.5 | — | 59.4 | — | — | — | — | |
| Diffusion-E2E-FTTraining Data=74K, Zero-shot=true2025.09 | 16.5 | — | 60.4 | — | — | — | — | |
| DSINE2026.05 | 16.5 | — | 58.75 | — | — | 9.17 | — | |
| IVGT2026.05 | 16.6 | — | — | 77.3 | 84.2 | 10.4 | — | |
| MUSE-dGenerative Model=×, MTL=✓, Inference Steps=12026.06 | 16.6 | — | 58.6 | — | — | — | — | |
| Lotus-dGenerative Model=×, MTL=×, Inference Steps=12026.06 | 16.8 | — | 58.2 | — | — | — | — | |
| Lotus-GInference Mode=Zero-shot2025.12 | 16.9 | — | 59.1 | — | — | — | — | |
| Lotus-gGenerative Model=✓, MTL=×, Inference Steps=12026.06 | 16.9 | — | 59.1 | — | — | — | — | |
| Lotus-2Generative Model=×, MTL=×, Inference Steps=12026.06 | 16.9 | — | 59 | — | — | — | — | |
| Lotus-2Method category=Generation-based2026.07 | 16.9 | — | 59 | — | — | — | — | |
| GeoWizardGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 17 | — | 56.5 | — | — | — | — | |
| MarigoldGenerative Model=✓, MTL=×, Inference Steps=12026.06 | 17.1 | — | 58.5 | — | — | — | — | |
| MUSE-gGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 17.1 | — | 58 | — | — | — | — | |
| Omnidata v22026.05 | 17.2 | — | — | 76.5 | 83 | 9.7 | — | |
| Omnidata V2Training Data=12.2M, Zero-shot=true2026.06 | 17.2 | — | 55.5 | — | 83 | — | — | |
| Lotus2026.05 | 17.25 | — | 58 | — | — | 9.43 | — | |
| StableNormal2026.05 | 17.55 | — | 55.9 | — | — | 10.26 | — | |
| OmniGen2Inference Mode=Zero-shot2025.12 | 17.7 | — | 54.6 | — | — | — | — | |
| StableNormalInference Mode=Zero-shot2025.12 | 17.8 | — | 54.2 | — | — | — | — | |
| Vision Banana2026.07 | 17.8 | — | — | — | — | — | — | |
| GenPerceptTraining Data=74K, Zero-shot=true2025.09 | 18.2 | — | 56.3 | — | — | — | — | |
| DiceptionGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 18.3 | — | 52.5 | — | — | — | — | |
| DICEPTIONMethod category=Generation-based2026.07 | 18.3 | — | 52.9 | — | — | — | — | |
| StableNormalTraining Data=250K, Zero-shot=true2025.09 | 18.6 | — | 53.5 | — | — | — | — | |
| StableNormalTraining Data=250K, Zero-shot=true2026.06 | 18.6 | — | 53.5 | — | 81.7 | — | — | |
| JoDiTraining Data=290K, Zero-shot=true2026.06 | 18.6 | — | — | — | — | — | — | |
| GeoWizardTraining Data=280K, Zero-shot=true2025.09 | 18.9 | — | 50.7 | — | — | — | — | |
| GeoWizardTraining Data=280K, Zero-shot=true2026.06 | 18.9 | — | 50.7 | — | 81.5 | — | — | |
| GeoWizardInference Mode=Zero-shot2025.12 | 19 | — | 50 | — | — | — | — | |
| DINOv2Backbone=g/142026.04 | 20.7 | — | — | — | — | — | — | |
| MarigoldTraining Data=74K, Zero-shot=true2025.09 | 20.9 | — | 50.5 | — | — | — | — | |
| MarigoldTraining Data=74K, Zero-shot=true2026.06 | 20.9 | — | 50.5 | — | — | — | — | |
| MarigoldMethod category=Generation-based2026.07 | 20.9 | — | 50.5 | — | — | — | — | |
| TIPSv2Backbone=g/142026.04 | 21.7 | — | — | — | — | — | — | |
| TIPSBackbone=g/142026.04 | 21.9 | — | — | — | — | — | — | |
| CORE-MTL2026.06 | 22.4927 | — | 35.01 | 61.96 | 73.6 | 16.8337 | — | |
| SigLIP2Backbone=SO/142026.04 | 23 | — | — | — | — | — | — | |
| FairGrad2026.06 | 23.0717 | — | 36.57 | 62.39 | 73.12 | 16.3569 | — | |
| OmnidataTraining Data=12.2M, Zero-shot=true2026.06 | 23.1 | — | 45.8 | — | 73.6 | — | — | |
| STCH2026.06 | 23.2045 | — | 36.1 | 62.03 | 72.86 | 16.5197 | — | |
| MTAN2026.06 | 24.0165 | — | 34.62 | 60.23 | 71.3 | 17.3356 | — | |
| RLW2026.06 | 24.0588 | — | 34.65 | 60.33 | 71.34 | 17.289 | — | |
| MOML2026.06 | 24.0679 | — | 34.69 | 59.87 | 70.93 | 17.413 | — | |
| IndividualBackbone=ResNet-502024.12 | 24.2 | — | — | — | — | — | — | |
| Single Task2026.06 | 24.2676 | — | 30.71 | 58 | 70.48 | 18.6778 | — | |
| Equal Weighting2026.06 | 24.2909 | — | 33.93 | 59.27 | 70.51 | 17.7635 | — | |
| CLIPBackbone=L/142026.04 | 24.3 | — | — | — | — | — | — | |
| PCGrad2026.06 | 24.3858 | — | 33.98 | 59.15 | 70.39 | 17.7999 | — | |
| GradNorm2026.06 | 24.3864 | — | 34.25 | 59.44 | 70.55 | 17.6391 | — | |
| RepMTL2026.06 | 24.5348 | — | 30.18 | 55.76 | 68.47 | 19.6139 | — | |
| ExcessMTL2026.06 | 26.1446 | — | 30.43 | 54.99 | 66.76 | 19.8459 | — | |
| SyMergeBackbone=ResNet-502024.12 | 26.2 | — | — | — | — | — | — | |
| EMR-MergingBackbone=ResNet-502024.12 | 26.5 | — | — | — | — | — | — | |
| LiNeS w/ TABackbone=ResNet-502024.12 | 29.1 | — | — | — | — | — | — | |
| OASIS2026.05 | 29.2 | — | — | 48.4 | 60.7 | 23.4 | — | |
| Weight AveragingBackbone=ResNet-502024.12 | 30 | — | — | — | — | — | — | |
| MagMaxBackbone=ResNet-502024.12 | 30.3 | — | — | — | — | — | — | |
| Task ArithmeticBackbone=ResNet-502024.12 | 30.6 | — | — | — | — | — | — | |
| Surgery w/ TABackbone=ResNet-502024.12 | 34.7 | — | — | — | — | — | — | |
| Ties-MergingBackbone=ResNet-502024.12 | 36.2 | — | — | — | — | — | — | |
| ProbSurgery w/ TABackbone=ResNet-502024.12 | 36.7 | — | — | — | — | — | — | |
| AnyUpResolution=224x2242025.10 | — | 31.17 | 29 | 57 | 69 | — | — | |
| AnyUp2026.06 | — | 27.83 | 49.62 | 70.14 | 77.67 | — | — | |
| BilinearResolution=224x2242025.10 | — | 32.7 | 26 | 53 | 66 | — | — | |
| Bilinear2026.06 | — | 28.23 | 49.33 | 69.74 | 77.18 | — | — |