Maxwell Render

The technique is called Photometric Stereo. The basic idea is you have a camera and a light, and you take 3 or more photos of the surface with the light in different positions. Using lambert's equation i = pN.L you can determine the normal of every pixel in the image.

If you take more than 3 photos the system becomes overdetermined, and you can start compensating for outliers like shadows and specularity.

The biggest strength of the method is that its per pixel. Got a 24MP camera? Then you have a scan with 24mil verts, the detail is great. The downside is that it breaks down for low frequency positional data. For example, scanning faces will result in some poor results for the overall shape. This is because the normal map generated from this method needs to be integrated into a height field. Since normal maps lack true depth, such as sharp discontinuities, low freq shapes get distorted.

BUT, medium frequency details and high frequency features like scars, skin pores, hair will all be incredibly crisp and accurate. If you're painting depthmaps, in my experience, the positional data isnt really needed (or wanted), bas relief data is easier to work with in mudbox and zbrush, so its a non-issue.

Also, in addition to a normal map, and depth, you get a flat shaded diffuse map. I know that artists painting face models love models with ambient lighting to reduce specularity and shading, and thats exactly what you get here.

All summed up, the method has its flaws, but its a perfect fit for painting bas relief high-res geometric details in apps like zbrush/mudbox.

Do you guys have any ideas about some sort of surfaces that could test out the limits of his system or to find shortcomings? I'll pass along one or two suggestions for him to try and bring back the results. Not necessarily anything that has extreme displacement, because it looks like this setup can definitely handle that, but just some things that it might be hard to create an accurate displacement for.