Boosting Monocular Depth Estimation Models to High-Resolution via Content-Adaptive Multi-Resolution Merging
Neural networks have shown great abilities in estimating depth from a single image. However, the inferred depth maps are well below one-megapixel resolution and often lack fine-grained details, which limits their practicality. Our method builds on our analysis on how the input resolution and the scene structure affects depth estimation performance. We demonstrate that there is a trade-off between a consistent scene structure and the high-frequency details, and merge low- and high-resolution estimations to take advantage of this duality using a simple depth merging network. We present a double estimation method that improves the whole-image depth estimation and a patch selection method that adds local details to the final result. We demonstrate that by merging estimations at different resolutions with changing context, we can generate multi-megapixel depth maps with a high level of detail using a pre-trained model.
PDF Abstract CVPR 2021 PDF CVPR 2021 AbstractCode
Colab
Datasets
Middlebury 2014
IBims-1
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Monocular Depth Estimation | IBims-1 | Miangoleh et al. (SGR) | ORD | 0.3938 | # 1 | |
| D3R | 0.3222 | # 1 | ||||
| RMSE | 0.1598 | # 1 | ||||
| δ1.25 | 0.6390 | # 1 | ||||
| Monocular Depth Estimation | IBims-1 | Miangoleh et al. (MiDaS) | ORD | 0.5538 | # 2 | |
| D3R | 0.4671 | # 2 | ||||
| RMSE | 0.1965 | # 2 | ||||
| δ1.25 | 0.7460 | # 2 | ||||
| Monocular Depth Estimation | Middlebury 2014 | Miangoleh et al. (MiDaS) | ORD | 0.3467 | # 1 | |
| D3R | 0.1578 | # 1 | ||||
| RMSE | 0.1557 | # 1 | ||||
| δ1.25 | 0.7406 | # 1 | ||||
| Monocular Depth Estimation | Middlebury 2014 | Miangoleh et al. (SGR) | ORD | 0.3879 | # 2 | |
| D3R | 0.2324 | # 2 | ||||
| RMSE | 0.1973 | # 2 | ||||
| δ1.25 | 0.7891 | # 2 |
