Bug in bicubic upscale in Brox OF (Bug #1821)
Description
Bicubic upscale in NPP_staging.cu has a bug.
How to reproduce: take a couple of 1360x1000 images, set eta to 0.8, number of solver iterations to 10, number of inner iterations to 5, number of outer iterations to 150. This should produce a wide band of NaN in optical flow field.
I guess the best solution is replace call to resize from NPP_staging with a call to corresponding OpenCV function.
But removing division by wsum (NPP_staging.cu:6798), will fix the problem.
History
Updated by Marina Kolpakova almost 13 years ago
- Assignee set to Marina Kolpakova
Updated by Marina Kolpakova almost 13 years ago
Thanks for reporting!
- Status changed from Open to Done
- Target version set to 2.4.0
- Category set to gpu (cuda)