In this work, we use unsupervised depth estimators to predict depth maps and optimize the distances between predicted depth maps and miscalibrated projected depth images.
Estimator:
- Check out this branch Monodepth[1]
We provide several upsampling methods implements for calibration and depth refinement after calibration. We also modified some upsampling methods for less time cost and higher accuracy.
| Method | RGB | Linear | Nearest | KNN | Barycentric | Grid Weight |
|---|---|---|---|---|---|---|
| Result |  |
 |
 |
 |
 |
 |
| Method | Anisotropic Diffusion[2] | SD-Filter[3] | Geometry | Modified Spatial[4] | Total Generalized Variation[5] | Clough Tocher |
|---|---|---|---|---|---|---|
| Result |  |
 |
 |
 |
 |
 |
We use Hyperopt for based optimization algorithms because the projection and cost calculation process is not explicitly convex and differentiable. An approximate training detail for three rotation parameters:
| Method | Bayesian Optimization | Stimulated Anneal |
|---|---|---|
| Loss |  |
 |
| RGB | Predicted depth map | Calibrated depth map |
|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
[1] Unsupervised Monocular Depth Estimation with Left-Right Consistency. Clément Godard, Oisin Mac Aodha and Gabriel J. Brostow. In CVPR, 2017
[2] Guided Anisotropic Diffusion and Iterative learning for Weakly Supervised Change Detection. Rodrigo Caye Daudt, Bertrand Le Saux, Alexandre Boulch, Yann Gousseau. In CVPR, 2019
[3] Robust Image Filtering using Joint Static and Dynamic Guidance. Bumsub Ham and Minsu Cho and Jean Ponce. In CVPR, 2015
[4] A Novel Way to Organize 3D LiDAR Point Cloud as 2D Depth Map Height Map and Surface Normal Map. Yuhang He, Long Chen, Jianda Chen, Ming Li. In ROBIO, 2015
[5] Image Guided Depth Upsampling using Anisotropic Total Generalized Variation. David Ferstl, Christian Reinbacher, Rene Ranftl, Matthias Rüther and Horst Bischof. In ICCV, 2013
Our work is under Creative Commons Legal Code.