Closed theartful closed 2 weeks ago
Yes, the current implementation of the underlying data structures of the reconstruction are not thread safe. Triangulating the observations in one image may affect the observations in other images.
I see, thank you for your responses. I try running glomap on videos, and on some of these videos the retrinagulation steps takes more than 75% of the time. Do you think it's worthwhile for me to look into it, or is there something fundamental preventing it.
Is there any technical reason preventing this section of the code from being parallelized? The retriangulation / bundle adjustment loop takes a lot of time in my tests. Specifically the retriangulation section.
https://github.com/colmap/glomap/blob/4210f9392567c4f38dfd1300db0126b69606fc0e/glomap/controllers/track_retriangulation.cc#L62-L70