Photometric Stereo - Sock

Ali_Youssef

Howdy counters,
I've been doing some tests for photometric stereo for a project I'm currently working on. Inspired by Dave and Gökhan, who showed great results using this technique.
Of course being a poor student, I had to improvise a setup that literally costs nothing, and I'm here to show the results which I'm quite happy about.
If anyone's interested about the daft setup that I got, I'll be happy to talk about that.

I wanted to try something that shines light through the surface, because my goal is to start scanning leaves and branches and various other plants, so I chose a sock because it's both interesting surface-wise (fabric) *and* shines a bit of light through, so I'd be testing detailed normal maps and SS. (in other words, it was the closest thing I could find without getting out of my chair).

Anyway, less talky, more showy:




Closeup:


Subsurface:


And finally, a gif of subsurface in action:


This was processed in Substance Designer and rendered in UE4. (flat plane with displacement)
All maps are in 8k, just to see what it would look like.

Two things that weren't really great about this, some areas are shadowed in all 8 original pictures, so there was no clean image of those areas for designer to generate albedo from, so they just look blurred and desaturated (apparent in the first image).
And secondly, I would definitely not scan dark objects like half of this sock, designer thought these were shadows and the heightmap (and normal map) got artifacts because of this, albedo on the other hand looks fine.

Nonetheless, I think it would look great as a decal or summat


Thanks for stopping by !

Shyralon

Hey, the results look pretty good already!
I am super interested in this technique (I even wrote my bachelor thesis about how to scan vegetation with this technique) and would love to hear more about your process. Especially your "cheap" scan setup

Ali_Youssef

Shyralon said:

Hey, the results look pretty good already!
I am super interested in this technique (I even wrote my bachelor thesis about how to scan vegetation with this technique) and would love to hear more about your process. Especially your "cheap" scan setup

Thanks Ramon !
I remember reading an article you wrote on 80 level, you got some pretty neat-looking assets out of it.
Do you have any advice with the technique ? I'm about to delve into some foliage production and could use some dos and don'ts.

As for my setup, it consist entirely of my fairly cheap phone, two chairs, a glass pane and a desk lamp.
I put the object on the glass to get the translucency map, and hold the lamp and move it around. The chairs are for holding the glass pane above the ground.

A problem I used to have is shaking the phone when pressing the "capture" button, so I DIYed a remote control using an xbox 360 controller plugged into my pc, which controlled my phone via teamviewer. A very backwards method but it works, so eh..

Shyralon

Unfortunately no, I am pretty much at the beginning myself and trying to figure out a good way to get satisfying results with cheap equipment. My setup was pretty much the same except for the remote control idea. Had the same problem with the shaking phone tho, so I might try your way as well
Have you ever used Dabarti Capture instead of Substance?
From what I can tell it is not necessarily better, but does different things very well (and it outputs a true heightmap which is cool)

Ali_Youssef

Shyralon said:

Unfortunately no, I am pretty much at the beginning myself and trying to figure out a good way to get satisfying results with cheap equipment. My setup was pretty much the same except for the remote control idea. Had the same problem with the shaking phone tho, so I might try your way as well
Have you ever used Dabarti Capture instead of Substance?
From what I can tell it is not necessarily better, but does different things very well (and it outputs a true heightmap which is cool)

I have read a bit about Dabarti Capture, however I knew I was still going to generate the other maps in substance, so I figured there would be no point in adding another software to the workflow.
The heightmap is definitely an issue, and even in this simple test, generating the heightmap from normals made the darker areas flatter than bright ones, which is understandable considering the algorithm, so it still needs manual fixing.
But I think for foliage it might not be too big of an issue, so I'm satisfied enough to not have the need to try Dabarti yet.