Affine-invariant Depth Estimation on ETH3D (AbsRel, delta1)
3.8AbsRelFE2E
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FE2Emode=zero-shot2025.09 | 3.8 | 98.7 | |
| FE2ETraining Data=71K, Pre-trained Stable Diffusion=false2025.09 | 3.8 | 98.7 | |
| FE2EGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 3.8 | 98.7 | |
| Ours CSTM_label (ViT-Large)Backbone=ViT-Large, #Params=306M, Pretrain=142M, #Data Train=16M, Zero-shot transfer=true2024.03 | 4.2 | 98.7 | |
| Ours CSTM_label (ViT-giant)Backbone=ViT-giant, #Params=1011M, Pretrain=142M, #Data Train=16M, Zero-shot transfer=true2024.03 | 4.2 | 98.3 | |
| Metric3Dv2Generative Model=×, MTL=✓, Inference Steps=12026.06 | 4.2 | 98.3 | |
| Lotus-2Generative Model=×, MTL=×, Inference Steps=12026.06 | 4.6 | 97.6 | |
| DiceptionGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 5.3 | 96.7 | |
| DINOv3mode=zero-shot2025.09 | 5.4 | 97.5 | |
| IrisTraining Data=59K + 100K, Zero-shot=true2026.03 | 5.5 | 97.6 | |
| Iris (ours)*Training Data=59K + 100K, diffusion-based=true2026.03 | 5.5 | 97.6 | |
| Lotus-G*Training Data=59K, Zero-shot=true2026.03 | 5.9 | 97 | |
| Lotus-G*Training Data=59K, diffusion-based=true2026.03 | 5.9 | 97 | |
| Lotus-GTraining Data=59K, Pre-trained Stable Diffusion=true2025.09 | 5.9 | 97 | |
| MUSE-dGenerative Model=×, MTL=✓, Inference Steps=12026.06 | 5.9 | 97.1 | |
| Lotus-D*Training Data=59K, Zero-shot=true2026.03 | 6.1 | 97 | |
| Lotus-D*Training Data=59K, diffusion-based=true2026.03 | 6.1 | 97 | |
| Lotus-DTraining Data=59K, Pre-trained Stable Diffusion=true2025.09 | 6.1 | 97 | |
| Ours CSTM_label (ViT-Small)Backbone=ViT-Small, #Params=22M, Pretrain=142M, #Data Train=16M, Zero-shot transfer=true2024.03 | 6.2 | 95.5 | |
| Lotus-gGenerative Model=✓, MTL=×, Inference Steps=12026.06 | 6.2 | 96.1 | |
| Ours CSTM_image (ConvNeXt-Large)Backbone=ConvNeXt-Large, #Params=198M, Pretrain=14.2M, #Data Train=8M, Zero-shot transfer=true2024.03 | 6.4 | 96.5 | |
| Diffusion-E2E-FT*Training Data=74K, Zero-shot=true2026.03 | 6.4 | 95.9 | |
| Diffusion-E2E-FT*Training Data=74K, diffusion-based=true2026.03 | 6.4 | 95.9 | |
| Diffusion-E2E-FTTraining Data=74K, Pre-trained Stable Diffusion=true2025.09 | 6.4 | 95.9 | |
| GeoWizardGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 6.4 | 96.1 | |
| MarigoldBackbone=Latent diffusion V2, #Params=899M, Pretrain=5B, #Data Train=74K, Zero-shot transfer=true2024.03 | 6.5 | 96 | |
| Marigold*Training Data=74K, Zero-shot=true2026.03 | 6.5 | 95.9 | |
| Marigold*Training Data=74K, diffusion-based=true2026.03 | 6.5 | 95.9 | |
| MarigoldTraining Data=74K, Pre-trained Stable Diffusion=true2025.09 | 6.5 | 95.9 | |
| Ours CSTM_label (ConvNeXt-Large)Backbone=ConvNeXt-Large, #Params=198M, Pretrain=14.2M, #Data Train=8M, Zero-shot transfer=true2024.03 | 6.6 | 96.4 | |
| GeoWizard*Training Data=280K, Zero-shot=true2026.03 | 6.6 | 95.8 | |
| GeoWizard*Training Data=280K, diffusion-based=true2026.03 | 6.6 | 95.8 | |
| Qwen-Imagemode=zero-shot2025.09 | 6.6 | 96.2 | |
| GeoWizardTraining Data=280K, Pre-trained Stable Diffusion=false2025.09 | 6.6 | 95.8 | |
| Marigold(LCM)*Training Data=74K, Zero-shot=true2026.03 | 6.8 | 95.6 | |
| Marigold(LCM)*Training Data=74K, diffusion-based=true2026.03 | 6.8 | 95.6 | |
| Lotus-dGenerative Model=×, MTL=×, Inference Steps=12026.06 | 6.8 | 95.3 | |
| GenPerceptTraining Data=74K, Pre-trained Stable Diffusion=false2025.09 | 7 | 95.6 | |
| Marigold v1.1Training Data=74K, Pre-trained Stable Diffusion=true2025.09 | 7 | 95.5 | |
| Marigoldv1.1Generative Model=✓, MTL=×, Inference Steps=12026.06 | 7 | 95.5 | |
| MarigoldGenerative Model=✓, MTL=×, Inference Steps=12026.06 | 7.1 | 95.1 | |
| OrchidGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 7.3 | 96.9 | |
| DPT-largeBackbone=ViT-Large, Pretrain=1.3M, #Data Train=188K, Zero-shot transfer=true2024.03 | 7.8 | 94.6 | |
| DPTTraining Data=1.4M, Zero-shot=true2026.03 | 7.8 | 94.6 | |
| DPTTraining Data=1.4M2026.03 | 7.8 | 94.6 | |
| MUSE-gGenerative Model=✓, MTL=✓, Inference Steps=12026.06 | 8.8 | 96.2 | |
| GenPercept*Training Data=90K, Zero-shot=true2026.03 | 9.4 | 96.1 | |
| GenPercept*Training Data=90K, diffusion-based=true2026.03 | 9.4 | 96.1 | |
| HDNBackbone=ViT-Large, #Params=306M, Pretrain=1.3M, #Data Train=300K, Zero-shot transfer=true2024.03 | 10.8 | 83.3 | |
| HDNTraining Data=300K, Zero-shot=true2026.03 | 12.1 | 83.3 | |
| HDNTraining Data=300K2026.03 | 12.1 | 83.3 | |
| DepthAnythingBackbone=ViT-Large, #Params=142M, Pretrain=142M, #Data Train=63.5M, Zero-shot transfer=true2024.03 | 12.7 | 88.2 | |
| DepthAnythingTraining Data=62.6M, Zero-shot=true2026.03 | 12.7 | 88.2 | |
| DepthAnythingTraining Data=62.6M2026.03 | 12.7 | 88.2 | |
| DepthAnythingTraining Data=62.6M, Pre-trained Stable Diffusion=false2025.09 | 12.7 | 88.2 | |
| DepthAnything V2Training Data=62.6M, Zero-shot=true2026.03 | 13.1 | 86.5 | |
| DepthAnything V2Training Data=62.6M2026.03 | 13.1 | 86.5 | |
| DepthAnything V2Training Data=62.6M, Pre-trained Stable Diffusion=false2025.09 | 13.1 | 86.5 | |
| OmnidataBackbone=ViT-Base, Pretrain=1.3M, #Data Train=12.2M, Zero-shot transfer=true2024.03 | 16.6 | 77.8 | |
| OmnidataTraining Data=12.2M, Zero-shot=true2026.03 | 16.6 | 77.8 | |
| OmnidataTraining Data=12.2M2026.03 | 16.6 | 77.8 | |
| LeresBackbone=ResNeXt101, Pretrain=1.3M, #Data Train=354K, Zero-shot transfer=true2024.03 | 17.1 | 77.7 | |
| LeResTraining Data=354K, Zero-shot=true2026.03 | 17.1 | 77.7 | |
| LeResTraining Data=354K2026.03 | 17.1 | 77.7 | |
| MiDaSBackbone=ResNeXt101, #Params=88M, Pretrain=1.3M, #Data Train=2M, Zero-shot transfer=true2024.03 | 18.4 | 75.2 | |
| MiDaSTraining Data=2M, Zero-shot=true2026.03 | 18.4 | 75.2 | |
| MiDaSTraining Data=2M2026.03 | 18.4 | 75.2 | |
| MiDaSTraining Data=2M, Pre-trained Stable Diffusion=false2025.09 | 18.4 | 75.2 | |
| DiverseDepthBackbone=ResNeXt50, #Params=25M, Pretrain=1.3M, #Data Train=320K, Zero-shot transfer=true2024.03 | 22.8 | 69.4 | |
| DiverseDepthTraining Data=320K, Zero-shot=true2026.03 | 22.8 | 69.4 | |
| DiverseDepthTraining Data=320K2026.03 | 22.8 | 69.4 |