Depth Prediction on ETH3D
4.8AbsRelDiT-based Diffusion for Perceptual Tasks
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DiT-based Diffusion for Perceptual TasksBackbone=DiT-XL2024.11 | 4.8 | 97.8 | — | — | |
| Marigold2024.11 | 6.5 | 96 | — | — | |
| DPT2024.11 | 7.8 | 94.6 | — | — | |
| DPT-LargeTraining set=MIX 6, Zero-shot cross-dataset transfer=true2021.03 | 8.9 | — | — | — | |
| DPT-HybridTraining set=MIX 6, Zero-shot cross-dataset transfer=true2021.03 | 9.3 | — | — | — | |
| MiDaSTraining set=MIX 6, Zero-shot cross-dataset transfer=true2021.03 | 11.6 | — | — | — | |
| Hierarchical Depth NormalizationTraining Data=~ 300K, Zero-shot transfer=true2022.10 | 12.1 | 83.3 | — | — | |
| HDN2024.11 | 12.1 | 83.3 | — | — | |
| MiDaSTraining set=MIX 5, Zero-shot cross-dataset transfer=true2021.03 | 12.9 | — | — | — | |
| GLDzero-shot=true2026.03 | 16 | 80 | 41 | 14.8 | |
| Omnidata2024.11 | 16.6 | 77.8 | — | — | |
| LeresTraining Data=~ 360K, Zero-shot transfer=true2022.10 | 17.1 | 77.7 | — | — | |
| LeReS2024.11 | 17.1 | 77.7 | — | — | |
| LiTraining set=MD [22], Zero-shot cross-dataset transfer=true2021.03 | 18.1 | — | — | — | |
| LiTraining set=MC [21], Zero-shot cross-dataset transfer=true2021.03 | 18.3 | — | — | — | |
| MiDaSTraining Data=~ 2M, Zero-shot transfer=true2022.10 | 18.4 | 75.2 | — | — | |
| MiDaS2024.11 | 18.4 | 75.2 | — | — | |
| XianTraining set=RW [45], Zero-shot cross-dataset transfer=true2021.03 | 18.6 | — | — | — | |
| Matrix3D2026.03 | 19.7 | 73.1 | 47.5 | 14.13 | |
| WangTraining set=WS [40], Zero-shot cross-dataset transfer=true2021.03 | 20.5 | — | — | — | |
| DiverseDepthTraining Data=~ 300K, Zero-shot transfer=true2022.10 | 22.8 | 69.4 | — | — | |
| DiverseDepth2024.11 | 22.8 | 69.4 | — | — | |
| CasserTraining set=CS [8], Zero-shot cross-dataset transfer=true2021.03 | 23.5 | — | — | — | |
| Chen et al.Training Data=~ 795K, Zero-shot transfer=true2022.10 | 23.7 | 67.2 | — | — | |
| WSVDTraining Data=~ 1.5M, Zero-shot transfer=true2022.10 | 26.1 | 61.9 | — | — | |
| Xian et al.Training Data=~ 30K, Zero-shot transfer=true2022.10 | 27.3 | 63 | — | — | |
| OASISTraining Data=~ 140K, Zero-shot transfer=true2022.10 | 29.2 | 59.5 | — | — | |
| OursBackbone=ResNeXt101, Zero-shot=true2020.12 | 31.9 | 77.7 | — | — | |
| OursBackbone=ResNet50, Zero-shot=true2020.12 | 34.4 | 75.8 | — | — | |
| WSVDBackbone=ResNet50, Zero-shot=true2020.12 | 35.9 | 61.9 | — | — | |
| ChenBackbone=ResNet50, Zero-shot=true2020.12 | 38.4 | 67.2 | — | — | |
| DiverseDepthBackbone=ResNeXt50, Zero-shot=true2020.12 | 38.6 | 69.4 | — | — | |
| MegaDepthBackbone=Hourglass, Zero-shot=true2020.12 | 39.8 | 64.3 | — | — | |
| MegaDepthTraining Data=~ 150K, Zero-shot transfer=true2022.10 | 39.8 | 52.7 | — | — | |
| MiDaSBackbone=ResNeXt101, Zero-shot=true2020.12 | 40.5 | 75.2 | — | — | |
| XianBackbone=ResNet50, Zero-shot=true2020.12 | 52.6 | 63 | — | — | |
| OASISBackbone=ResNet50, Zero-shot=true2020.12 | 60.2 | 59.5 | — | — |