Capturing Real World Surfaces for PBR

McBradd · Mar 2014

After pouring over the Quixel Megascans page, I want to learn the process behind capturing real-world surfaces.

"...each pixel of every map has been physically scanned, and every material component has been fully separated. Every reflectance value has been measured, and each normal reveals how light interacts with the surface and subsurface."

I'm familiar with capturing diffuse, and using polarizing filters to capture and derive Specular contribution, but that is just careful use of a DSLR. Creating Normal Maps from photos is kind of hit or miss from there, and generally not the best practice.

They hint at some proprietary technical wizardry, [ame="

http://www.youtube.com/watch?v=8alYZgkwClM"]without coming right out and claiming so.[/ame] Pretty sweet Quixel branded PCB though.

TL;DR
My question is:
How do you capture (rather than derive) these maps?

Specular
Gloss
Normal
Displacement
Ambient Occlusion
Translucency
Scatter
Transparency

Are we talking about some crazy laser/radar/holographic projection/scanning tech, or just clever use of a nice camera?

To anyone with experience capturing these materials that can guide me in the right direction, Thanks.

passerby · Mar 2014

they are prolly projecting light from different directions, and seeing what it hits to get data similar to normals

ZacD · Mar 2014

Here's a way to do photos -> normal maps
http://www.zarria.net/nrmphoto/nrmphoto.html

SHEPEIRO · Mar 2014

[ame="

http://www.youtube.com/watch?v=_wIgmEch1P8"]Ready At Dawn Textile Scanning Acquisition - YouTube[/ame]
may not be the same but it gives you quite a good idea of how ready at dawn are doing it

Joost · Mar 2014

I think they probably use cross-polarization for the specular. EDIT: oops I can't read, sorry

Hopefully they'll go into depth a bit more when megascans is released. I'd really like to construct a DIY rig to produce my own (crappy) megascans.

cman2k · Mar 2014

There's a lot of bits and pieces out there, but I have never seen anyone do a complete breakdown and I've always been interested in digging in myself. Here's some relevant links:

Creating accurate normals from photos has been around for a while;
http://www.zarria.net/nrmphoto/nrmphoto.html

Splitting specular and diffuse let's you measure the specular intensity and know how much light is being reflected.
http://filmicgames.com/archives/233

Fox engine tech video talks about creation of linear textures from photos;
http://imgur.com/a/08z8k or https://www.youtube.com/watch?v=17nje72VnPE (the full talk)

if you think of a "reverse" process for SSbump (like normal maps) you can imagine using the right lighting setup to capture ambient occlusion from photos in the same way you capture normal maps (possibly using harsh angled lighting vs soft angled lighting and getting the difference?)
https://developer.valvesoftware.com/wiki/$ssbump

In the Quixel video you can see that they lined the inside of their box with LEDs - and I'm guessing they are doing that so they can tightly control the lighting situation from many different angles easily. When you split specular and diffuse lighting - you are really just taking data from one lighting situation and another and pulling out the difference. It seems to me that you could use very similar techniques for stuff like AO, translucency, etc.

cman2k · Mar 2014

to accompany Ready At Dawn's youtube above;

lots of details here! they pretty much break down their whole setup, but it's just for albedo/normals. they are using generalized BDRFs for the reflectance of stuff like cloth, rather than trying to measure it, I think.
course notes; http://blog.selfshadow.com/publications/s2013-shading-course/rad/s2013_pbs_rad_notes.pdf

some slides from their powerpoint: http://imgur.com/a/DYv30

full slides; http://blog.selfshadow.com/publications/s2013-shading-course/rad/s2013_pbs_rad_slides.pptx

I think this pretty much confirms that this is possible with a very simple, low-cost setup (excepting the camera itself) - and no crazy lasers are required!

Vaiktorg · Dec 2016

In the video, the dude is holding a Circular Polarizer filter and a Neutral Density Filter. Specifically:
- Neutral Density: Light Craft Workshop 77mm Fader ND MKII ND Filter LCW
- Circular Polarizer: HOYA 77mm CIR-PL Slim

Joopson · Dec 2016

cman2k said:
Creating accurate normals from photos has been around for a while;
http://www.zarria.net/nrmphoto/nrmphoto.html

I know this is from a post from 2014, but every time I see that link shared, I can't help but cringe a little. It's a very clever technique to photosource a very very rough approximation of normal information, but it's not "accurate" since it contains shadow information. The shading will always be bizarre on anything that produces any cast shadows. And it's very hard to remove cast shadows in the source photos. Again, very very clever technique, but the results won't be true to their source.

Capturing Real World Surfaces for PBR

Replies