Metric Depth Estimation on NYU Metric Depth v2 (test)
98.9Delta 1 AccuracyMetric3D v2
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Metric3D v2Backbone=ViT-L, CSTM=label, Protocol=fine-tuned2024.03 | 98.9 | 99.8 | 100 | 0.047 | 0.02 | 0.183 | |
| Metric3D v2Backbone=ViT-g, CSTM=label, Protocol=fine-tuned2024.03 | 98.7 | 99.7 | 99.9 | 0.045 | 0.015 | 0.187 | |
| DepthAnythingProtocol=fine-tuned2024.03 | 98.4 | 99.8 | 100 | 0.056 | 0.024 | 0.206 | |
| UniDepthType=I → M, Scaling=Zero-shot, Train Images=3M, Train Params=347M2025.06 | 98.1 | 99.7 | 100 | 0.072 | 0.03 | 0.229 | |
| Metric3D v2Backbone=ViT-g, CSTM=label, Protocol=zero-shot2024.03 | 98 | 99.7 | 99.9 | 0.067 | 0.03 | 0.26 | |
| Metric3Dv2Type=I → M, Scaling=Zero-shot, Train Images=16M, Train Params=1011M2025.06 | 98 | 99.7 | 99.9 | 0.067 | 0.03 | 0.26 | |
| Metric3D v2Backbone=ViT-L, CSTM=label, Protocol=zero-shot2024.03 | 97.5 | 99.4 | 99.8 | 0.063 | 0.028 | 0.251 | |
| PolymaxProtocol=zero-shot2024.03 | 96.9 | 99.6 | 99.9 | 0.067 | 0.029 | 0.25 | |
| DA V2Type=I → M, Scaling=Domain Adap, Train Params=25M2025.06 | 96.9 | 99.6 | 99.9 | 0.073 | 0.032 | 0.261 | |
| DAType=R → M, Scaling=Linear Fit, Train Images=64M, Train Params=25M2025.06 | 96.9 | 99.5 | 99.9 | 0.055 | 0.024 | 0.26 | |
| UniK3DType=I → M, Scaling=Zero-shot, Train Images=8M, Train Params=359M2025.06 | 95.8 | 99.1 | 99.6 | 0.072 | 0.031 | 0.268 | |
| DAType=I → M, Scaling=Domain Adap, Train Params=25M2025.06 | 95.7 | 99.5 | 99.9 | 0.075 | 0.033 | 0.273 | |
| ZeroDepthProtocol=fine-tuned2024.03 | 95.4 | 99.5 | 100 | 0.074 | 0.031 | 0.269 | |
| TR2MType=(I,L) R → M, Scaling=Rescale maps, Train Images=102K, Train Params=19M2025.06 | 95.4 | 99.6 | 99.9 | 0.082 | 0.035 | 0.293 | |
| ZoeDepthProtocol=fine-tuned2024.03 | 95.3 | 99.5 | 99.9 | 0.077 | 0.033 | 0.277 | |
| ZoeDepthType=I → M, Scaling=Domain Adap, Train Params=345M2025.06 | 95.1 | 99.4 | 99.9 | 0.077 | 0.033 | 0.282 | |
| Metric3D v2Backbone=ConvNext-L, CSTM=label, Protocol=zero-shot2024.03 | 94.4 | 98.6 | 99.5 | 0.083 | 0.035 | 0.31 | |
| IEBins2024.03 | 93.6 | 99.2 | 99.8 | 0.087 | 0.038 | 0.314 | |
| WorDepthType=I,L → M, Scaling=Domain Adap, Train Params=137M2025.06 | 92.6 | 99 | 99.8 | 0.09 | 0.04 | 0.33 | |
| Metric3D v2Backbone=ConvNext-L, CSTM=image, Protocol=zero-shot2024.03 | 92.5 | 98.3 | 99.4 | 0.092 | 0.04 | 0.341 | |
| NeWCRFs2024.03 | 92.2 | 99.2 | 99.8 | 0.095 | 0.041 | 0.334 | |
| ScaleDepthType=I,L → M, Scaling=Domain Adap, Train Params=109M2025.06 | 91.3 | 98.9 | 99.8 | 0.099 | 0.041 | 0.329 | |
| Adabins2024.03 | 90.3 | 98.4 | 99.7 | 0.103 | 0.044 | 0.364 | |
| ZeroDepthProtocol=zero-shot2024.03 | 90.1 | 96.1 | — | 0.1 | — | 0.38 | |
| TrDepth2024.03 | 90 | 98.3 | 99.6 | 0.106 | 0.045 | 0.365 | |
| VNL2024.03 | 87.5 | 97.6 | 99.4 | 0.108 | 0.048 | 0.416 | |
| DepthLMType=I,L → M, Scaling=Zero-shot2025.06 | 82.4 | 93.3 | 99 | 0.134 | 0.068 | 0.401 | |
| Laina et al.2024.03 | 81.1 | 95.3 | 98.8 | 0.127 | 0.055 | 0.573 | |
| Li et al.2024.03 | 78.8 | 95.8 | 99.1 | 0.143 | 0.063 | 0.635 | |
| RSAType=(I,L) R → M, Scaling=Rescale factors, Train Images=102K, Train Params=4.7M2025.06 | 75.2 | 96.4 | 99.2 | 0.156 | 0.071 | 0.528 | |
| DAType=R → M, Scaling=Global, Train Images=64M, Train Params=25M2025.06 | 63 | 92.6 | 98.7 | 0.199 | 0.087 | 0.646 | |
| DAType=R → M, Scaling=Median, Train Images=64M, Train Params=25M2025.06 | 48 | 73.4 | 88.6 | 0.353 | 0.135 | 1.743 | |
| DepthCLIPType=I,L → M, Scaling=Zero-shot2025.06 | 39.4 | 68.3 | 85.1 | 0.388 | 0.156 | 1.167 |