search

google-research / multinerf

A Code Release for Mip-NeRF 360, Ref-NeRF, and RawNeRF
Apache License 2.0
3.58k stars 339 forks source link

How to apply a different tonemapping algorithm? #31

Open Harper714 opened 1 year ago

Harper714 commented 1 year ago

Hi, if I want to get the result of HDR tonemapping, I can change exposure and the parameters in linear_to_srgb, right? Is there any advice to change these parameters?

jonbarron commented 1 year ago

Yeah I guess you could directly change that function to have a different shape, but it's used during training so I'm not sure what effect that will have. The easiest thing to do is to train a rawnerf, render a linear intensity image, and then apply your own tone mapping function to the linear rendering that you get out.

Harper714 commented 1 year ago

Thanks for the answer.

BTW, I trained the raw model for the scene parkstatue using default configs, and rendered the views by setting Config.factor =1. Then I noticed that when zoom-in the rendered image, i.e., color_000.png, there are obvious stripe artifacts on the sky region near the lamp. I think the possible reason is related to the Bayer mosaic mask but I am not sure. Have you noticed this issue, and what do you think the reason is?

bmild commented 1 year ago

I think that in scenes such as parkstatue, the noise level is so high that some artifacts are hard to avoid, either as a result of COLMAP error or from the NeRF optimization.

This is the color_000.png output I see from the last run I did for parkstatue. Are these similar artifacts to the ones you're getting? image