ByteDance-Seed / ByteDance-Seed/Depth-Anything-3
Severe depth estimation degradation with no-background samples
Open
- Dominant language
- Python
- Stars
- 6.3k
- Forks
- 702
- PR merge metrics
- No merged PRs in 30d
Description
The depth estimation capability of the NESTED-GIANT-LARGE model degrades significantly on no-background samples (e.x. pure black or white background), which affects camera pose estimation and 3D object generation.
In addition, I found that the depth estimation of DA3 has degradation on the surfaces of transparent and reflective objects, as well as on some fine structures.
I wonder if anyone has encountered the same problem as I have.
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.