I am trying to figure out the best way to use RC to simulate an incremental Sfm and reconstruction workflow. I am interested in exploring if this software can be used in a scenario that is low-latency/quasi-real time. I’ve a question on component alignments and also I wonder if anybody here has any suggestions or alternate ways of approaching this problem. Question: When importing and aligning components, does RC take the model (3D mesh) along with the component? Or, does RC need to do reconstruction again after components have been aligned together? I only have the demo version of RC, so can’t try this out via the GUI - will save me time and or $$ if I can answer this question here instead of via direct experimentation.
I think the best approach may be to incrementally align small components and do model reconstruction from sub-sets of data. The data would be previously ordered in this simulation so images are added in adjacent blocks. Then, after building the new component, I would align it to the already built model, aggregating to larger components or a single model. In this fashion, I could incrementally build up a large model but not incur a huge processing delay at any single step. Overall time is probably going to be longer than if I did a single, global solution, but I have my reasons for needing to do things incrementally.
From what I can follow in the help files and forum discussions, it should technically be possible to do this kind of incremental build using the cache. I’ve experimented with this via the GUI, by building a component and model. Then I locked the poses of the aligned cameras and added a couple unregistered cameras. Sure enough the alignment step saved time, but when I did the reconstruction, there was no time savings. As far as I can tell the model was recalculated from scratch, discarding the previous results. This behavior is what has motivated my original question. Perhaps this was because I could not save and import the exact some reconstruction region? Are there specific conditions that need to be satisfied to get RC to re-use model results already stored in the cache?
Any hints/comments/suggestions are appreciated!