Problems with -aa 0 near edges

radiance-dev
1

Folks,

In experimenting with the Turrell piece "Afrum I (White)" I found what
seems to be a bug in the interpolation procedure that becomes notable
for -aa 0.

Image: http://markjstock.org/transfer/img12.png
Files: http://markjstock.org/transfer/afrum1.tar

This is with HEAD from today. See the HOWTO file for command-line.

Mark

2

Very strange, I'll grant you. If you turn off super-sampling (-as 0) or increase -ad to 3000, the artifact mostly disappears. My suspicion is that the method by which the ambient calc estimates errors to decide where to send its super-samples has some regularities due to the hemispherical subdivision I use. Near the inside corners, the visible hemisphere is split into two regions, one of which is much brighter than the other. Moving a small amount left or right causes a large number of divisions to cross over the threshold, which shifts the super-samples and therefore the bias in the calculation.

This bug falls under the category of general shortcomings of the strategies used in ambient sampling, which I hope to largely replace with the Hessian calculation I mentioned in an earlier post.

Cheers,
-Greg

ยทยทยท

From: Mark Stock <[email protected]>
Date: September 3, 2013 7:46:43 PM PDT

Folks,

In experimenting with the Turrell piece "Afrum I (White)" I found what
seems to be a bug in the interpolation procedure that becomes notable
for -aa 0.

Image: http://markjstock.org/transfer/img12.png
Files: http://markjstock.org/transfer/afrum1.tar

This is with HEAD from today. See the HOWTO file for command-line.

Mark