search

CVLAB-Unibo / neural-disparity-refinement

Other
46 stars 4 forks source link

About SGM and AD-Census #3

Closed tholmb closed 2 years ago

tholmb commented 2 years ago

You mentioned in the paper that you were generating on the fly noisy disparity maps computed by two traditional stereo algorithms SGM and AD-Census. With SGM do you mean OpenCV's SGBM (cv::StereoSGBM)?

If I'm correct OpenCV do not provide either AD-Census algorithm but it can be done for example based on this thread https://stackoverflow.com/questions/38265364/census-transform-in-python-opencv. You were talking in the supplementary material about AD-Census's parameters a uniqueness ratio and block size and I was thinking that did you used (cv::StereoBM) as AD-Census?

Anyways great work!

tholmb commented 2 years ago

Just noticed that OpenCV actually contains algorithm for SGM (https://docs.opencv.org/5.x/d1/ddc/classcv_1_1cuda_1_1StereoSGM.html). I will keep digging if I can also find direct opencv function for AD-Census.

tholmb commented 2 years ago

Answer to myself: You was probably using your own c++ implementation https://github.com/fabiotosi92/Unsupervised-Confidence-Measures/blob/master/README.md.