Hi all. I am working with a small group of people, porting bundler, which recovers the camera positions and 3D points from a collection of uncalibrated images, to VXL. We were hoping to make the design very modular, so that people can either use the general, all-purpose approach currently implemented, or specify different constraints by implementing their own subroutines to plug into the bundler pipeline. Currently the bundler source code is somewhat scattered and difficult to use.
We had a couple of questions, I guess the first of which is, is anyone interested in this? Currently, bundler is one of the best implementations of the structure from motion problem, so we think that adding it to a mature vision library might allow for cool applications, but we're wondering about the general interest level.
Secondly, we were thinking of adding this to the "contrib" folder; would this be the correct place to put it? We briefly considered trying to add it to vil_algo, but we felt that if the project was successful and people were interested, we might migrate it to that directory after we finished and people had some time to use it. If we do add it to the contrib directory, what name would be the most consistent?
Finally, is anyone interested in critiquing the proposed architecture? I don't want to flood the list with technical details if no one cares, but we would like to here thoughts and opinions on what we're planning to do from people very familiar with VXL.