Does anyone have any thoughts on using photogramettry for game characters? I find lots of documentation of using it for Static Meshes but little to none for game characters. I assume this is because the polycount on scanned meshes is often too high to cope with the demands of rigging and animation? Would anyone be able to recommend a maximum poly count for game characters scanned this way? Keeping in mind there will be 100+ in the final scene. Realism is not a goal so I’m happy to go very low, would just be good to get an idea what people feel is required.
Kojima production used photoscan on Metal Gear Solid V
In their presentation of the Fox engine, they also show how they used photoscan on characters and other assets.
To reduce the high polycount, the details were baked down to normal maps, using Simplygon.
The high res photoscans provided source for various material elements.
Photoscans are not perfect, you still need to clean the meshes up and process the Textures. So if realism is not your goal then using photoscan might not be worth all of the required clean up work to get it game ready.
Another reason photogrammetry isn’t used on characters as much is budgetary. To scan a static object, all you need is one camera and good lighting conditions. People move, however, which means either your actor/model has to stay perfectly still, or you need to create a multi-camera rig (to get the scan done in a fraction of a second) which can get expensive very quickly.
Sure, but you still need to organize that and pay for it. Lots of people have a DSLR just lying around. I think we’re in agreement that there’s no significant technical/workflow reason not to use photogrammetry for game characters. I was just trying to shed light for hmihalis1 on a few of the hurdles you have to get past in order to do it.
I actually use Photogrammetry for all my Characters, to some extent anyway. I usually only do hands for photogrammetry in order to capture all the micro-detail correctly, everything else I usually sculpt. Even though I usually use my own scans I’ve also purchased a few pre-cleaned scans from Ten24. You might want to check them out, even though it’s a bit expensive as a result of USD currency conversion (Their store site uses Euros.)
But anyway, I personally tend to avoid facial captures because they are often too difficult to get acceptable results from a single camera’s work. Hands are easier to work with certainly. And I personally make sure to only one finger per chunk and then merge the resulting chunks by dense point cloud alignment in Agisoft. I usually get pretty decent results. With that being said, I’m also fortunate enough to have set aside a particular space in my workshop for Photogrammetry and tuned the lighting conditions properly.
As far as polycount goes, what I usually tend to do and create a low-res mesh in Fuse that roughly fits the sillhoutte of the photogrammetry hand. And then I reproject the high-res detail onto the new mesh, export it with vertex color baked in, and then bake out all the maps in Substance Painter. I could certainly make a video of my workflow if people are interested.