I'm trying to speed up a fur rendering.
For the most parts of the object 5x5 pixel samples will work fine, but at the edges of the object, where the fur becomes thinner, 9x9 samples are needed.
My idea is to use an greyscale image as multiplier for the pixel sampling. The map could be created rendering ultra low res alpha channels - contract and smooth and invert the image, so that there is black where the inner object part is and grey to white where the outline or rim of the object is. Black = 5x5 samples - grey 6x6 - 8x8 samples - white 9x9 samples.
That's the plan.
I'm pretty new to houdini and do not know how to get values from an image nor if or how houdini can handle the grey values for every render tile due it needs to calculate the sum of digits for every tile to get only one value as sampling multiplier per tile. Even after googeling for hours i did not found an howto use image maps as values/multipliers.
5x5 samples ~6 minutes rendertime per frame
9x9 samples ~21 minutes rendertime per frame
Thats why i would like to get this kinda adaptive sampling. Or is there another solution for my problem?