How to get the stereo depth maps?

[yavon818]

Thanks for your great work! I tried to run the program on my own data, however, I dont know how to get the stereo depth maps of the stereo images. I remembered that you mentioned using the DERS tool, but I dont know how to get access to it. Could you give me some hints?

Thanks a lot!

[jlartois]

Hi, there are many ways to calculate depth maps. Here are some suggestions:

1. DERS is relatively old, a more current version is called IVDE . I find it quite slow though, so I would not recommend it anymore.
2. If you have a pair of undistorted stereo images, you can estimate their disparity, and thus the depth using OpenCV.
3. If you have multiple images, and not necessarily stereo pairs, then you should take a look at Multi-View Stereo (MVS). The best MVS algorithms generally are based upon PatchMatch.
4. COLMAP allows you to generate depth maps during the dense reconstruction..
5. You could train NeRF (or the faster instant-ngp) on your images and render the depth maps.
6. Lately, many deep learning models are trained to estimate depth maps from a set of input images. These methods are fast but might not generalise well for every type of image. Some examples: MVSNet, MultiViewStereoNet

So if you want to learn more about MVS depth estimation, I would recommend to look into PatchMatch based approaches.