| Image | Normals RGB | Normals Needle | Albedo | Depths | Depths w/Albedo |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Overall the results came out surprisingly well. My lighting directions come out slightly different then the sample program; however the discrepancy is not reflected in any discernable way in the results. The parts of the images that didn't turn out so well were the owl's eye and the gray sphere. The owl’s eye was so dark that under all of the lighting conditions it appeared black which caused the spiky noise in both the sample and my application. The grey sphere ends up looking less round than I would expect it to be. It may be the case that the sphere actually is distorted as seen in the 3-D representation however from looking at the input images it looks much rounder to me. As the algorithm is currently designed you also end up with some peculiar artifacts around the edges of the image. Improvements could include detecting the shiny surfaces and then smoothing those areas to reduce the noise. The artifacts at the edges of the model could likely just be dropped by filtering out parts of the model that were sparse.