yqueau
commented
5 years ago HI,
It actually depends how you tune the \lambda parameter.
- If \lambda goes to zero the prior z0 matters less and less, it basically only enforces well-posedness of the problem and you won't observe any difference between using an estimated depth map or initializing it with zeros
- if \lambda goes to infinity then you will force the solution to be the same as z0
- inbetween the result will mix the depth prior with the normals information. Can be useful for instance if you have a noisy depth map to denoise with not-so-accurate normals, or to create some funny mix between two shapes. But tuning \lambda might be a bit tricky, as it will basically be just trial and error.
Hope this helps !
Le jeu. 14 févr. 2019 à 15:57, abeacco notifications@github.com a écrit :
Hi,
It's me again, sorry :P
I am wondering if there is a specific format for the depth prior z0 to feed your algorithms, as I am not seeing any difference between using an estimated depth map information or initializing it to zeros. Is it because the algorithm gets always to the same solution?
Thanks again for this great work :)
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/yqueau/normal_integration/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AKB5rILThigIJMBp4UlW9otGANYEYUL7ks5vNXlQgaJpZM4a7xPo .
abeacco
Hi,
It's me again, sorry :P
I am wondering if there is a specific format for the depth prior z0 to feed your algorithms, as I am not seeing any difference between using an estimated depth map information or initializing it to zeros. Is it because the algorithm gets always to the same solution?
Thanks again for this great work :)