Gaussian Blur Kernel Size and Sigma

[pietz]

I haven't validated this in detail because I don't run the code in this repo. I only use it for looking up settings. I noticed that the point masks are blurred with a Gaussian kernel size 15 and a sigma of 4. When I use the same settings in OpenCV this will result in cropped Gaussians because 15 isn't large enough for a sigma of 4. I'm using 25.

Just wanted to bring this up as a potential problem. Not sure if it really is.