search

Kai-46 / IRON

Inverse rendering by optimizing neural SDF and materials from photometric images
BSD 2-Clause "Simplified" License
299 stars 24 forks source link

Why should we use "roughness range loss" #4

Closed Woolseyyy closed 2 years ago

Woolseyyy commented 2 years ago

Hi Kai, Appreciate for your another great work! The results of IRON look amazing! However, I am confused about the "roughness range loss" mentioned in the paper. Why should we encourage the estimated roughness to stay below 0.5? What if we meet object with large roughness in custom data? Looking forward to your advice.

Kai-46 commented 2 years ago

The MitsubaroughplasticBRDF model we are using is already pretty diffuse (hardly any visible specularity) when roughness=0.5; hence we thought it might be better to constrain the values to be below 0.5.

Woolseyyy commented 2 years ago

Got it! Thank you!