Monocular Depth Estimation on RTX3090 480x640 efficiency (test)
3.1Latency (ms)Angular Module
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Angular ModuleBackbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU, Note=Evaluated independently2025.03 | 3.1 | — | 12.1 | |
| Camera ModuleHardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 5.1 | — | 13.4 | |
| Radial ModuleBackbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU, Note=Evaluated independently2025.03 | 21.9 | — | 38.2 | |
| BTSBackbone=D161 (DenseNet-161), Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 28.5 | 35.1 | 47 | |
| AdainsBackbone=EN-B5 (EfficientNet-B5-AP), Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 33.2 | 30.1 | 78.3 | |
| Metric3DBackbone=CNXT-L (ConvNext-L), Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 40.3 | 24.8 | 203.2 | |
| Depth ModuleHardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 49.2 | — | 26.6 | |
| NewCRFBackbone=SWin-L, Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 53.1 | 18.8 | 280 | |
| DepthAnything v2Input Size=518 x 518, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 78.1 | — | 334.7 | |
| iDiscBackbone=SWin-L, Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 81.1 | 12.3 | 209.2 | |
| UniDepthBackbone=CNXT-L (ConvNext-L), Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 86.6 | 11.5 | 238.9 | |
| UniK3DInput Size=518 x 518, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 88.4 | — | 358.8 | |
| Metric3Dv2Input Size=518 x 518, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 135.6 | — | 441.9 | |
| ZoeDepthBackbone=BEiT-L, Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 144.8 | 6.91 | 345.9 | |
| ZoeDepthInput Size=512 x 512, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 144.8 | — | 345.9 | |
| UniDepthBackbone=ViT-L, Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 146.4 | 6.83 | 347 | |
| UniDepthInput Size=518 x 518, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 146.4 | — | 347 | |
| MASt3RInput Size=512 x 512, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 154.7 | — | 668.6 | |
| DepthProInput Size=1536 x 1536, Backbone=ViT-L, Precision=16-bit float, Hardware=RTX3090 GPU2025.03 | 808.1 | — | 952 | |
| ZeroDepthBackbone=R18 (ResNet-18), Hardware=RTX3090, Precision=32-bit, Input Resolution=480x6402024.03 | 955.6 | 1.05 | 232.6 |