I am using PCL and need some help for setting the appropriate constant values while computing SIFT KeyPoints - minimum contrast, number of octaves, scales per octave. From the sift code, I understood that minimum scale is the voxel/leaf size, but other constants are not clear. The original research paper by Lowe on SIFT KeyPoints is based on image processing and I'm still trying to contextualize that for point cloud processing. I don't have any background in image processing. All the open source examples on the internet give some fixed values to these constants, but I think that they should be a function of the input point cloud. Any suggestions for the same? I also searched for some reference material on the internet to understand the nexus between images and point cloud (in the context of SIFT KeyPoints), but haven't found any thus far.
constexpr float minimumContrast = 1e-3f; constexpr int numberOctaves = 3, scalesPerOctave = 4;
These are the commonly used values, but I don't understand the logic behind using those values. I even get a result after setting these values, but I'd like to tune them so as to extract the optimal keypoints.