I was wondering if anyone has tried this: implementing a piece of sofrwae that takes n 2D non-occluded wireframe raster images as input and outputs a 3D polygon-based model that could be projected to each of the 2D images.

Actually forget raster because -- although not trivial -- I think it's within the state of the art to do image processing on a raster image of a solid repesented as a wireframe to extract a graph drawing that represents the wireframe image. So let's say the 2D to 3D algorithm take n sets of graphs plus 2D positions for the vertices as input.

Has someone written this? Googling around I'm only seeing commercial products that claim to go from photographs to textured models which I find hard to believe actually work (except maybe on the images that the developers have spent a lot of time getting to work) But, I think wireframes should be doable.

I almost worked on a project that did this. I resigned that job but it's certainly doable (resigned for personal reasons).
That was some time ago, but I found these stuff during my initial research.

http://ebookbrowse.com/l14-morestereo-pdf-d138179162
http://cbcl.mit.edu/people/poggio/journals/marr-poggio-PRSL-1979.pdf
http://www.vision.ca...papers/Tsai.pdf

i hope this helps.

Thanks for the links.

See what I'm thinking about is that I have a big collection Zome tools, which are like tinker toys on steroids. I think that photographs of zome constructions against a black background could be processed into graph drawings. I was hoping that a 2D wireframe to 3D model library was already out there somewhere, but it doesn't look like it.