Depth Estimation on DIODE (test)
0.065AbsRelDepth Anything V2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Depth Anything V2Encoder=ViT-G, Zero-shot=true2024.06 | 0.065 | 95.4 | |
| Depth Anything V1Encoder=ViT-L, Zero-shot=true2024.06 | 0.066 | 95.2 | |
| Depth Anything V2Encoder=ViT-L, Zero-shot=true2024.06 | 0.066 | 95.2 | |
| Depth Anything V2Encoder=ViT-B, Zero-shot=true2024.06 | 0.068 | 95 | |
| Depth Anything V1Encoder=ViT-B, Zero-shot=true2024.06 | 0.069 | 94.6 | |
| Depth Anything V2Encoder=ViT-S, Zero-shot=true2024.06 | 0.073 | 94.2 | |
| MiDaS V3.1Encoder=ViT-L, Zero-shot=true2024.06 | 0.075 | 94.2 | |
| Depth Anything V1Encoder=ViT-S, Zero-shot=true2024.06 | 0.076 | 93.9 | |
| DPT-largeBackbone=ViT-Large, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.182 | 75.8 | |
| DPTZero-shot=true, # Training samples=1.2M + 188K2024.12 | 0.182 | 75.8 | |
| CSTM_imageBackbone=ConvNeXt-large, #Params=198M, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.211 | 82.5 | |
| CSTM_labelBackbone=ConvNeXt-large, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.224 | 80.5 | |
| HDNBackbone=ViT-Large, #Params=306M, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.246 | 78 | |
| HDNZero-shot=true, # Training samples=300K2024.12 | 0.246 | 78 | |
| CrossFlowZero-shot=true, # Training samples=74K2024.12 | 0.27 | 76.8 | |
| LeresBackbone=ResNeXt101, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.271 | 76.6 | |
| LeReSZero-shot=true, # Training samples=300K + 54K2024.12 | 0.271 | 76.6 | |
| MarigoldZero-shot=true, # Training samples=74K2024.12 | 0.31 | 77.2 | |
| MiDaSBackbone=ResNeXt101, #Params=88M, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.332 | 71.5 | |
| MiDaSZero-shot=true, # Training samples=2M2024.12 | 0.332 | 71.5 | |
| OmnidataBackbone=ViT-base, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.339 | 74.2 | |
| OmnidataZero-shot=true, # Training samples=11.9M + 310K2024.12 | 0.339 | 74.2 | |
| DiverseDepthBackbone=ResNeXt50, #Params=25M, Zero-shot transfer=true, Manually aligned scale and shift=true2023.07 | 0.376 | 63.1 | |
| DiverseDepthZero-shot=true, # Training samples=320K2024.12 | 0.376 | 63.1 | |
| 4M-21Model Scale=XL, Evaluation Protocol=Out-of-the-box2024.06 | 0.68 | — | |
| 4M-21Model Scale=Large, Evaluation Protocol=Out-of-the-box2024.06 | 0.69 | — | |
| 4M-21Model Scale=Base, Evaluation Protocol=Out-of-the-box2024.06 | 0.71 | — | |
| Unified-IOModel Scale=XL, Evaluation Protocol=Out-of-the-box2024.06 | 0.82 | — | |
| Unified-IO 2Model Scale=XXL, Evaluation Protocol=Out-of-the-box2024.06 | 0.84 | — | |
| Unified-IO 2Model Scale=XL, Evaluation Protocol=Out-of-the-box2024.06 | 0.86 | — | |
| Unified-IOModel Scale=Large, Evaluation Protocol=Out-of-the-box2024.06 | 0.87 | — | |
| Unified-IO 2Model Scale=Large, Evaluation Protocol=Out-of-the-box2024.06 | 0.96 | — | |
| Unified-IOModel Scale=Base, Evaluation Protocol=Out-of-the-box2024.06 | 1 | — |