Great Work! I am trying to use Unidepth as inference. However, the depth map that I am obtaining is very blockish. Perhaps, this has to do with patch tokenization? I have attached a depth map below.
The backbone I am using is "ViTL14". Is this the expected depth output? How do I avoid this?
Patch tokenization may have an impact, are you using .infer method or are you resizing the images yourself? A wrong image size may have this effect, too.
Hello
Great Work! I am trying to use Unidepth as inference. However, the depth map that I am obtaining is very blockish. Perhaps, this has to do with patch tokenization? I have attached a depth map below.
The backbone I am using is "ViTL14". Is this the expected depth output? How do I avoid this?
Thank you!