Photogrammetry Model Processing (mainly UV mapping pains)

Loooong time no post! :eek:

I’ve been getting into photogrammetry a bit lately, and Ive run into an issue getting a good workflow setup. I was wondering if anyone had better ideas/suggestions/experience!

So basically the initially generated texture/model from the 3D scanning software is a bit of a mess. Very high poly, which is easily tamed through the decimation modifier, but its also a mess of materials and auto-generated UV maps.
With this example Im working on you can see its 1.6 million polys to begin with. The UV map is even more of a mess (which is really what youd come to expect from something automatically generated) and with this model its split across 6 different texture maps. A handful to deal with…

[SUP][SUB]Thats a lot of detail… send help[/SUB][/SUP]

Ive decimated the model down to 600k, which is much more workable without too much loss in mesh detail. But once the mesh is sorted next comes the issue of cleaning up and getting workable UV unwrapping to edit/bake textures.

I’m seeing two potential approaches:
a) Hand-edit seams with the lower poly model so it unwraps nicely, and then bake the textures across from the original mesh. Then with a clean UV layout and texture its easy enough to edit and modify in Photoshop etc.

b) ‘Smart UV Project’ to quickly generate UVs and bake textures across from the original mesh so everything is on one texture, then texture paint in Blender to clean up the texture.

Ive tried approach a) so far, with decent results. But I worry about having to unwrap larger, more details objects (like the rest of the car). Its a time consuming and difficult process to work out the seams by hand but I cant see any other ways of doing it at this point. Using smart UV project does a pretty hack job of setting up decent UVs on objects this detailed and I cant work with the result of that in Photoshop.

[SUB][SUP]This took a few hours of seam work, not the best UV mapping but its workable[/SUP][/SUB]

[SUP][SUB]Smart UV project auto-unwrapping makes mostly a big mess [/SUB][/SUP]

Im yet to try option b), as I dont have a lot of experience with texture painting in Blender yet, so I’ll need to do some tutorials and see if those tools are as capable as I need it to be. But of course with b) the issue is also having really messy texture setups, which could leave the model less flexible to tweaking/modifying later on down the track. But it is a LOT faster in terms of unwrapping…

Does anyone have any suggestions or experience with this? Are there tools or techniques Im missing?

Which software are you using? I have been using Reality Capture (unbelievably good) and within the software you can simplify your poly count, as well as isolate the region to reconstruct. Also why do you have so many different texture maps?

As for blender, I have been just modeling super low poly even density quad cage, and then subd/shrinkwrap/smooth combo 2-3 times until I get a good medium poly with good topology, apply them, mark my own seams and unwrap that and then bake the colormap/normal from scan to medium poly. I usually export my texture from CR at 8192, and then in blender use either 4096, or lower depending on how large I need it in the frame. On my rig I can usually handle somewhere around 1.5 million polys for my scan .obj

Sorry should have mentioned, Im using AutoDesk Remake. Ive given a few other programs a try but havent got the same quality results. Im yet to try Reality Capture though, might have to give it a go. Remake seems pretty limited when it comes to options overall.

The texture maps are just what it spits out, Im not sure why but its not something you can work with from the raw output out of remake, Ive been re-baking the textures into one 8192x8192 texture and that seems to be large enough to preserve the details.

So youd suggest taking the retopology route? I havent ever tried that technique myself, but that should definitely solve the topology issues. How does your quad-cage poly count compare to the photogrammetry mesh?

For a boot I just did, the scan was ~800K, and the retopo about 14K verts. You could definitely go lower (or higher) here’s some screenshots. There’s a free trial version of CR, although you cannot export with it.