Over-smoothness for background normal prediction

baegwangbin / DSINE

[CVPR 2024 Oral] Rethinking Inductive Biases for Surface Normal Estimation

https://baegwangbin.github.io/DSINE/

Other

685 stars 28 forks source link

Over-smoothness for background normal prediction #16

Open FantasticOven2 opened 2 weeks ago

FantasticOven2 commented 2 weeks ago

Hello authors, thanks for this amazing work!

I ran DSINE with checkpoints/exp001_cvpr2024/dsine.pt on some custom data, and here's the result. It seems the normals for background stairs are over-smoothed. Do you encouter similar problem during experiements?

0_00001 0_00001