This project is a revamp of my old stereo matching research with CUDA. In this repository in rewriten them using modern CUDA along with Pytorch to demonstrate how to Pytoh can be
Learning project of classic stereo matching algorithms. Implementation done in PyTorch, CUDA and C++.
Aproaches:
- Winner takes all
- Dynamic Programming
- Semiglobal matching
Besides the references for stereo matching algorithms, it also servers as example of a PyTorch + C++ + CUDA + skbuild project.
Note: The algorithms aren't differentiable.
$ pip install .