Testing Unity's experimental De-Lighting tech

petemc

I tested Unity's de-lighting tool on a rock scan from a few weeks back, here are the resuts!
model

A bit about the process here: 
https://petemcnally.com/2017/07/26/afternoon-de-lighting/

petemc

petemc

jStins

This is really cool, thanks for sharing!

petemc

jStins said:

This is really cool, thanks for sharing!

Thanks Joel! There were a few errors in my bakes, so I'm running the tests again over the weekend

petemc

I'll post the results on the blog

Dash-POWER

Hi, I've tried also this interesting tool. With example maps which are included on Git it works really well but I've tried my test scan and I got not really good results in some parts on my scanned object. See attached screen below. The object was scanned by smartphone without Xrite color checker - homeless photoscan system .
The input maps for the tool are baked in xNormal and doesn't looks bad as input I think. Only color map has too dark areas and I understand this can't be delighted because there aren't any surface color information but on some areas it looks that light could be removed. Instead of this I'm getting color fragments (specially on the Lizard's head). I'm using +X -Y +Z axis orientation (tried also +++ but got almost same results).



If someone ever had similar problem I would be happy if someone could help. (I'm understanding that sources aren't made by proper way of photogrammetry. I just want to know why I'm getting these color fragments if because of the tool has some bug or my sources are too fantastic :-D. I'm using all maps which are necessary for the tool.
You can download sources here(only maps): Google drive download

Thank you for any reply.

petemc

You could try baking a position map, Substance can bake them and I think XNormal too. Could you post a photo of the object you scanned so we could see how the albedo behaves in real life?

Dash-POWER

I've already baked position map in SD but it has no change. I'm starting thinking that you should bake your model with fixed world orientation because my object space normals has different world axis then example maps. Tool has Switch Y/Z button but result is almost same as before. Also tried TGA instead of PNG. Position map is just additional map.



GIF: Before delighting.
New tga maps: download

petemc

I think your mask is wrong, try masking the lizard from the wood.

Dash-POWER

petemc said:

I think your mask is wrong, try masking the lizard from the wood.

Yes. This helped a lot. Result is much better then before. It has still some fragments but it's probably because of bad scan. Thank you for your tip.

petemc

Much better, nice work!