Creating a large number of 3D heads of real people FAST

Snowfly

What the title says.

My company needs to create 3D models of 35 faculty members for a client, but we're pressed for time and manpower for this particular project. We basically have a month.

Barring starting from scratch or stylizing/simplifying, what methods can we use to quicken the process?

I'm downloading the 123D Catch and am playing around with My3DScanner to see if they suit our needs. Any other solutions we can look into?

LoTekK

Is Kinect data high-enough resolution for your needs? I haven't played with it myself yet, but it could be one avenue to look into, though I don't know what sort of point cloud resolution you can get out of it.

edit:
Just found these two things:
http://blenderartists.org/forum/showthread.php?202543-Kinect-3D-Scanner-development
http://www.ros.org/wiki/openni/Contests/ROS%203D/RGBD-6D-SLAM

I don't have a Kinect at home, so I won't be able to try them out, but at the very least the first one appears more or less feature complete, judging from the snippets of the discussion I looked at.

Snowfly

Going the Kinect route would be great, since we get realtime feedback. I'll see if I can wrangle work into getting one. Taking the snapshots for the 3D stitching apps has been a pretty cumbersome experience so far...

Oh and the final result should be in the ballpark of <1000 tris, 512x512 diffuse only.
So we're not after crazy resolution. The heads just need to have a passing likeness at a couple of inches high at most on a 23" HD res screen.

LoTekK

Y'know, it occurred to me that you could make one base head to spec, unwrapped and all, then shrink wrap that to each head that you get from the 3d scans (whether via Kinect or otherwise), and bake the diffuse down from the scan as well (and then of course do some cleanup on the texture). That way you're not worrying about retopoing each and every head scan.

Abdullah

try facegen. may look simple but it surely helps you in your short time.

fatihG_

I would have probably made a base head, uv it so that I simply put a photo of a person straight on from the front, primarily the eyes, nose, mouth and forehead.
Edit it to more or less resemble the heads of the subjects.

Annalize the 35 people and maybe make 3-4 different base hairstyles (spikey, long, etc) and edit those where needed.

Modeling the base head should not take that long. One day, two max, with UV's. And another day or two for the hairstyles.

After that its all about tweaking the base meshes and taking a photo from the front and the side and merging them, liquify in photoshop to match the UV's.
In an 8 hour work day you could easily create 2 heads I think. 2,5-3 hours for editing the model and 1-1,5 hours for the textures.

If everything would go according to plan you would end up with roughly a week for refining the models/textures or adding some more hairstyles.

Anyway that's how I would plan it.

Lennyagony

I played around with 123d catch on some static objects, its ok but im not convinced its a time saver.

You end up having to go through LoTekK's process above with the addition of 30x35 photos and more than likely will end up having to smooth out the scan mesh before the shrinkwrap. It also doesn't work on hair. If you have reasonably ok ambient lighting for the shoot it can produce some nice diffuse maps, but they will need to be baked down to a UV'd mesh.

All in all its going to be faster to go with taking a few front/side shots and manually adjusting your base mesh to match.

poopipe

Abdullah wrote: »

try facegen. may look simple but it surely helps you in your short time.

another vote for facegen - you tend to get a bit of a baggy neck on the default head but that can be sorted with a morph target/blendshape.

you just bang 2 or 3 photos in, let it do it's stuff and you've got a head in a few minutes - complete with a load of expression targets iirc. also it's very cheap