Monocular Depth Estimation on Stanford2D3D (test)
97.27δ1 AccuracyDA2
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DA22026.02 | 97.27 | 0.0616 | 0.4261 | 98.96 | 99.36 | — | — | — | |
| One2SceneEvaluation Protocol=Finetune2026.02 | 96.95 | 0.0444 | — | 98.85 | 99.44 | — | — | — | |
| Ours2026.02 | 96.86 | 0.0675 | 0.4317 | 98.81 | 99.37 | — | — | — | |
| UniK3DTraining Set=Ours (26 sources), Evaluation Mode=Zero-shot2025.03 | 96.8 | 0.0801 | — | — | — | — | — | — | |
| DAP2026.02 | 95.64 | 0.0686 | 0.4662 | 98.71 | 99.34 | — | — | — | |
| Joint with layout and semanticssupervision=layout and semantics2022.02 | 95.4 | 0.068 | 0.264 | 99.2 | 99.8 | — | — | — | |
| VGGT-360Backbone=VGGT [40], Train → Test=Training-free2026.03 | 95.3 | 0.068 | — | 98.3 | 99.1 | — | — | — | |
| One2SceneEvaluation Protocol=Zero-shot2026.02 | 95.2 | 0.0675 | — | 98.53 | 99.3 | — | — | — | |
| VGGT-360Backbone=Fastvggt [31], Train → Test=Training-free2026.03 | 95.2 | 0.07 | — | 98.3 | 99.1 | — | — | — | |
| VGGT-360Backbone=π³ [42], Train → Test=Training-free2026.03 | 95.2 | 0.065 | — | 98.4 | 99.3 | — | — | — | |
| MultiPanoWise2026.02 | 94.51 | 0.0649 | 0.3892 | — | — | — | — | — | |
| PanDA2026.02 | 93.96 | 0.0879 | 0.5989 | 98.56 | 99.33 | — | — | — | |
| HUSH2026.02 | 93.84 | 0.0782 | 0.3332 | 98.49 | 99.29 | — | — | — | |
| BiFuse++Loss=Affine-Inv, Training Data=M-all, ST-all (p), Zero-shot=true2024.06 | 93.1 | 0.082 | — | 97.9 | 99.1 | — | — | — | |
| Depth AnywhereBackbone=Depth Anything [44], BiFuse++ [39], Train → Test=M+ → S2026.03 | 93 | 0.083 | — | 97.8 | 99 | — | — | — | |
| Depth AnywhereBackbone=Depth Anything [44], UniFuse [19], Train → Test=M+ → S2026.03 | 92.7 | 0.082 | — | 97.8 | 99 | — | — | — | |
| BiFuse++Loss=Affine-Inv, Training Data=M-all, SP-all (p), Zero-shot=true2024.06 | 92.6 | 0.086 | — | 97.9 | 99.1 | — | — | — | |
| UniFuseLoss=Affine-Inv, Training Data=M-all, ST-all (p), Zero-shot=true2024.06 | 92.4 | 0.086 | — | 97.7 | 99 | — | — | — | |
| EGFormerLoss=Affine-Inv, Training Data=M-all, ST-all (p), Zero-shot=true2024.06 | 92.3 | 0.086 | — | 97.6 | 99 | — | — | — | |
| BiFuse++Loss=Affine-Inv, Training Data=M-all, Zero-shot=true2024.06 | 92.1 | 0.09 | — | 97.6 | 99 | — | — | — | |
| UniFuseLoss=Affine-Inv, Training Data=M-all, SP-all (p), Zero-shot=true2024.06 | 92 | 0.09 | — | 97.8 | 99 | — | — | — | |
| HoHoNetLoss=Affine-Inv, Training Data=M-all, ST-all (p), Zero-shot=true2024.06 | 92 | 0.088 | — | 97.9 | 99.2 | — | — | — | |
| HRDFusePatch size/FoV=256 x 256 / 80°2023.03 | 91.4 | 0.0935 | 0.3106 | 97.98 | 99.27 | — | 0.1422 | 0.0508 | |
| BiFuse++Loss=BerHu, Training Data=M-all, Zero-shot=true2024.06 | 91.4 | 0.107 | — | 97.5 | 98.9 | — | — | — | |
| UniFuseLoss=Affine-Inv, Training Data=M-all, Zero-shot=true2024.06 | 91.4 | 0.09 | — | 97.6 | 99 | — | — | — | |
| BiFuse++Training Set=Matterport3D, Evaluation Mode=Zero-shot2025.03 | 91.4 | 0.107 | — | — | — | — | — | — | |
| HRDFuse2026.02 | 91.4 | 0.0935 | — | 97.98 | 99.27 | — | — | — | |
| BiFuse++Backbone=ResNet34 [15], Train → Test=M → S2026.03 | 91.4 | 0.107 | — | 97.5 | 98.9 | — | — | — | |
| UniFuseLoss=BerHu, Training Data=M-all, Zero-shot=true2024.06 | 91.3 | 0.094 | — | — | — | — | — | — | |
| UniFuseTraining Set=Matterport3D, Evaluation Mode=Zero-shot2025.03 | 91.3 | 0.0942 | — | — | — | — | — | — | |
| UniFuseBackbone=ResNet34 [34], Train → Test=M → S2026.03 | 91.3 | 0.094 | — | — | — | — | — | — | |
| Depth Anywhere2026.02 | 91 | 0.118 | — | 97.1 | 98.7 | — | — | — | |
| PaGeRSeparate indoor or outdoor prediction heads=true2026.05 | 90.94 | 10.94 | 45.43 | — | — | — | — | — | |
| DAPSeparate indoor or outdoor prediction heads=true2026.05 | 90.64 | 10.97 | 53.39 | — | — | — | — | — | |
| HoHoNetLoss=Affine-Inv, Training Data=M-all, Zero-shot=true2024.06 | 90.6 | 0.095 | — | 97.5 | 99.1 | — | — | — | |
| EGFormerLoss=Affine-Inv, Training Data=M-all, Zero-shot=true2024.06 | 90.6 | 0.098 | — | 97.2 | 98.9 | — | — | — | |
| SliceNetMethod architecture=re-evaluated pretrained model2022.02 | 90.59 | — | 0.3383 | 96.35 | 98.48 | 0.1715 | 0.0752 | — | |
| HoHoNet2022.08 | 90.54 | 0.1014 | 0.3834 | 96.93 | 98.86 | 0.2027 | 0.0668 | — | |
| HoHoNet2026.02 | 90.54 | 0.1014 | — | 96.93 | 98.86 | — | — | — | |
| SliceNetstatus=recalculated using open source models2022.08 | 90.38 | 0.0998 | 0.3728 | 96.23 | 98.43 | 0.1737 | 0.0765 | — | |
| GLPanoDepth(Complete)Method architecture=Full architecture2022.02 | 90.15 | — | 0.3493 | 97.39 | 99.01 | 0.1932 | 0.068 | — | |
| PanoDepthsynthesized views=3, cascade design=two-level2022.02 | 90.01 | 0.0972 | 0.3747 | 97.01 | 99 | — | — | — | |
| OmniFusion (2-iter)Patch size/FoV=256 x 256 / 80°2023.03 | 89.88 | 0.095 | 0.3474 | 97.69 | 99.24 | — | 0.1599 | 0.0491 | |
| HRDFusePatch size/FoV=128 x 128 / 80°2023.03 | 89.41 | 0.0984 | 0.3452 | 97.78 | 99.23 | — | 0.1465 | 0.053 | |
| PanoFormerBackbone=Transformer, #Params (M)=20.38, #FLOPs (G)=81.092024.03 | 88.74 | 0.1122 | 0.3945 | 95.84 | 98.59 | — | — | 0.0786 | |
| Elite360DBackbone=ResNet-34, #Params (M)=25.51, #FLOPs (G)=65.282024.03 | 88.72 | 0.1182 | 0.3756 | 96.84 | 98.92 | — | — | 0.0728 | |
| ACDNet2026.02 | 88.72 | 0.0984 | — | 97.04 | 98.95 | — | — | — | |
| Elite360D2026.02 | 88.72 | 0.1182 | — | 96.84 | 98.92 | — | — | — | |
| TaskPrompter2026.02 | 88.29 | 0.1171 | 0.5792 | 97.5 | 99.06 | — | — | — | |
| PanoFormer*Patch size/FoV=-/-2023.03 | 88.08 | 0.1131 | 0.3557 | 96.23 | 98.55 | — | 0.2454 | 0.0723 | |
| PanoFormer2026.02 | 88.08 | 0.1131 | — | 96.23 | 98.55 | — | — | — | |
| BridgeNet2026.02 | 87.97 | 0.1198 | 0.5834 | 97.38 | 98.92 | — | — | — | |
| BiFuse++2026.02 | 87.83 | — | — | 96.49 | 98.84 | — | — | — | |
| InvPT2026.02 | 87.75 | 0.1207 | 0.581 | 97.19 | 98.66 | — | — | — | |
| UniFuse with fusionPatch size/FoV=-/-2023.03 | 87.11 | 0.1114 | 0.3691 | 96.64 | 98.82 | — | — | — | |
| UniFuse2026.02 | 87.11 | 0.1114 | — | 96.64 | 98.82 | — | — | — | |
| UniFuseBackbone=ResNet-34, #Params (M)=50.48, #FLOPs (G)=96.522024.03 | 87.06 | 0.1124 | 0.3555 | 97.04 | 98.99 | — | — | 0.0709 | |
| OmniFusionBackbone=ResNet-34, #Params (M)=42.46, #FLOPs (G)=142.292024.03 | 86.74 | 0.1154 | 0.3809 | 96.03 | 98.71 | — | — | 0.0775 | |
| SphereDepth2022.08 | 86.66 | 0.1158 | 0.4512 | 96.42 | 98.63 | 0.2323 | 0.0754 | — | |
| BiFuse with fusion2022.02 | 86.6 | 0.1209 | 0.4142 | 95.8 | 98.6 | — | — | — | |
| BiFuseMethod architecture=standard2022.02 | 86.6 | — | 0.4142 | 95.8 | 98.6 | 0.2343 | 0.0787 | — | |
| BiFuse2022.08 | 86.6 | 0.1209 | 0.4142 | 95.8 | 98.6 | 0.2343 | 0.0787 | — | |
| BiFuse with fusionPatch size/FoV=-/-2023.03 | 86.6 | 0.1209 | 0.4142 | 95.8 | 98.6 | — | — | — | |
| BiFuse2026.02 | 86.6 | 0.1209 | — | 95.8 | 98.6 | — | — | — | |
| Dinh et al.2026.02 | 86.5 | 0.12 | 0.39 | 98.8 | 99.1 | — | — | — | |
| BiFuseLoss=BerHu, Training Data=M-all, Zero-shot=true2024.06 | 86.2 | 0.12 | — | — | — | — | — | — | |
| BiFuseTraining Set=Matterport3D, Evaluation Mode=Zero-shot2025.03 | 86.2 | 0.12 | — | — | — | — | — | — | |
| BiFuseBackbone=ResNet50 [15], Train → Test=M → S2026.03 | 86.2 | 0.12 | — | — | — | — | — | — | |
| DACBackbone=ResNet101 [15], Train → Test=In+ → S2026.03 | 85.9 | 0.124 | — | 97.6 | 99.1 | — | — | — | |
| RectNet2022.02 | 83.26 | 0.1409 | 0.4568 | 95.18 | 98.22 | — | — | — | |
| GLPanoDepth(Concat)Method architecture=Concatenation fusion2022.02 | 82.11 | — | 0.4829 | 94.16 | 97.62 | 0.2734 | 0.0906 | — | |
| EGFormerBackbone=Transformer, #Params (M)=15.39, #FLOPs (G)=66.212024.03 | 81.85 | 0.1528 | 0.4974 | 93.38 | 97.36 | — | — | 0.1408 | |
| EGFormer2026.02 | 81.85 | 0.1528 | — | 93.38 | 97.36 | — | — | — | |
| GLPanoDepth(ViT+CNN)Method architecture=ViT and CNN hybrid2022.02 | 81.72 | — | 0.481 | 93.86 | 97.73 | 0.2739 | 0.0901 | — | |
| DepthAnyCamera2026.05 | 77.98 | 17.15 | 59.29 | — | — | — | — | — | |
| FCRN2022.02 | 72.3 | 0.1837 | 0.5774 | 92.07 | 97.31 | — | — | — | |
| FCRNMethod architecture=standard2022.02 | 72.3 | — | 0.4142 | 92.07 | 97.31 | 0.3428 | 0.11 | — | |
| FCRN2022.08 | 72.3 | 0.1837 | 0.5774 | 92.07 | 97.31 | 0.3428 | 0.11 | — | |
| FCRNPatch size/FoV=-/-2023.03 | 72.3 | 0.1837 | 0.5774 | 92.07 | 97.31 | — | — | — | |
| MarigoldLoss=Affine-Inv, Training Data=Pers., Zero-shot=true2024.06 | 69.2 | 0.195 | — | 94.2 | 98.2 | — | — | — | |
| OmniDepthMethod architecture=standard2022.02 | 68.77 | — | 0.6152 | 88.91 | 95.78 | 0.3743 | 0.1212 | — | |
| OmniDepth2022.08 | 68.77 | 0.1996 | 0.6152 | 88.91 | 95.78 | 0.3743 | 0.1212 | — | |
| UniK3DIn-domain training=true2026.05 | 64.5 | 24.37 | 64.71 | — | — | — | — | — | |
| 360MDBackbone=MiDaS v2 [28], Train → Test=Training-free2026.03 | 63.6 | 0.268 | — | 87.8 | 94.5 | — | — | — | |
| Depth AnythingLoss=Affine-Inv, Training Data=Pers., Zero-shot=true2024.06 | 63.5 | 0.248 | — | 89.9 | 97 | — | — | — | |
| RPG3602026.05 | 17.36 | 29.35 | 90.73 | — | — | — | — | — |