The main problems we had were in the triangulation computations. There were some holes in the meshes. This was probably because the system had a hard time mapping a 3D coordinate to the very bright surface.
The other problem was with the sizes of certain pieces. Some of the views we had were either slightly bigger or slightly smaller that the other peices. This was probably due to some precision error when mapping the camera points to the object points. Perhaps the distortion factor wasn't as acccurate as we thought it would be.
We would have liked to scan in more views of the object. As we observed in zipper, reconstructing an object required a lot of overlapping views to get an accurate representation. Although the results we got were pretty good, it could have been a lot better with more view. One of the major problems with adding more view, however, is the increased computation time. Zipper was already quite slow while working with our object (it had about 42K triangles). Adding more views would have slowed it down to the point of being unusable (zipper did crash quite a few times while we were working with it). If we had more time, it would have been nice to scan the top and bottom view of the object. Also, texture mapping would have been nice.
Some things we would have done differently was to streamline the entire scanning process. Right now we had to do a lot of the file manipulation and processing by hand. It would have been easier to scan in a lot of images if we streamlines the process (through scripts, etc.). Also, I would have like to use multiprocessor machines or distibute the load across a network of machines. It seems that zipper would slow the machine down significantly.
Another interesting idea would be to see what the differences were with using different reconstruction techniques. For example, it would be nice if we can try scanning an object using two cameras instead of one camera and a structured light source. Then again, seeing how little time we had, this would have been near impossible. How about a class demo?
It is probably a good idea to have milestones for the class. Students should be required to present parts of their project at certain times. This will help in avoiding the end of the term crunch.