Scatter estimation has lower interpolation errors if linear interpolation is used

[markus-jehl]

On a simulated cylinder, the scatter estimation has much smaller errors when we use linear interpolation instead of quadratic interpolation. I strongly suspect that this is because we quite aggressively downsample the scanner (e.g. 8 instead of 48 rings) to speed up the computation time. Quadratic interpolation simply can't cope with that level of simplification. Below is the quadratic interpolation vs. ground truth first, then the linear interpolation vs. ground truth:

Originally posted by @markus-jehl in #1291 (comment)