I made this material to test of how much missing data can be reconstructed by Atomatix's Artificial Intelligence based on existing information.
As I wanted to create the material which covers roughly 1.8 by 1.8 square meters I had no enough highpoly surface data coverage to bake from. Also since FBX can handle vertex color information just for 67mln poly model and generated one was 99mln, I cut some geometry to fit the budget. It was caused without any quality loss as I removed redundant and non-consistent parts of the surface only - these were useless anyway. When baked I got textures with about 50% of missing data but also 50% highly detailed an.
I painted mask in Artomatix to mark areas to be ignored by AI and use whats left for reconstruction. As you can see it is almost 50% to 50% but the surface itself is quite generic and consistent.
I had no idea what result am I going to get and was surprised how well AI handled that. It took just 5 secs to AI to get it done nice and clean.

This is my photogrammetry workflow - still experimenting with 'Artomatix' and replaced Photoscan with Metashape (more details regarding to my workflow at: https://www.artstation.com/artwork/K0PYG).