While the test set is still not complete even with respect to covering
all shape types, I've better flushed out test cases for verifying the
inside test mechanism. In the process found some of our basic shapes are
not today returning the correct inside test results. Further, I've long
had on my list that some of the media intensities look suspect to me for
the non-inside test based versions and I'll show one example - blobs.
In the attached image master branch used at commit 054e75c except the
bottom row which uses the my new solver branch given how bad the blob
result for speckles was with the master branch in the next to bottom row.
The top row shows an expected result for shapes in general. In this case
for a bezier_spline prism. When things are working the media result in
the left column uses the shape straight up as a container with constant
media density. The middle column shows a media result using a matching
inside-test -> object-pattern -> function based media. Should be the two
approaches basically match though they won't be identical.
Rows one through four showing shapes where the inside test is failing.
The bottom two rows show a class of cases on my list for a while where
it looks to me like the normal shape-intersection media intensities are
too high. Not dug at all - bounding issue for the inside test version
maybe - don't know.
- Row 0 A good test.
- Row 1 the cylinder.
- Row 2 the cone (cone and cylinder really the same code internally).
- Row 3 superellipsoid set as a sphere (a).
- Row 4 the new torus-union variety is acting like the torus-merge for
the inside test.
- Row 5 blob test with master. Speckling bad. Shape OK but intensities
puzzling.
- Row 6 blob done test with new solver branch re-based off master.
(a) - Remember the peeling paint superellipsoid scene of Thomas's where
I tried to use the soft_object to create actual peeling paint on the
fly. Remember too how it was offset away from the surface of the
superellipsoid. I until yesterday though I had just gotten something
wrong in my offset math - looks instead like the dang inside test is
wrong (set for the largest box possible maybe)!
Anyway, more stuff for the todo list.
Bill P.
Post a reply to this message
Attachments:
Download 'insidetestissues.png' (214 KB)
Preview of image 'insidetestissues.png'
|