3d_photo_stylization icon indicating copy to clipboard operation
3d_photo_stylization copied to clipboard

Published 20 hours ago • verified publisher icon fmu2

I tried several depth model, and found the render result of MonoDepth is better than LeReS

Open lewizo opened this issue 2 years ago • 1 comments

The detail of LeReS is better than Mono. But because of the plentiful detail, pixels are more cluttered on the render result. Such as the resdual trace of hairs or the broken limbs. I think it's not normal. How can I use the depths of LeReS to be better?

https://user-images.githubusercontent.com/65765608/180351072-e81e1890-f203-47e8-869e-b8694b0652c8.mp4

https://user-images.githubusercontent.com/65765608/180351102-7e5261e8-e1e8-47dc-bd51-412dc6072f5a.mp4

lewizo avatar Jul 22 '22 02:07 lewizo

In general we found DPT works better than LeReS and earlier depth estimation methods. The strength of LeReS is that it additionally estimates camera parameters. To obtain clean results you may want to compare different depth estimation models and pick the one that gives sharp depth boundaries.

fmu2 avatar Sep 10 '22 19:09 fmu2