Thank you for the amazing work on PixSfM — it's an impressive system!
I have a question regarding the integration of GPS data into the reconstruction pipeline. Currently, the standard workflow is to:
- Run COLMAP(or PixSfM) to build the initial SfM model.
- Use model_aligner to align the model with GPS data via similarity transformation.
- Feed the aligned model into PixSfM for refinement.
Would it be possible (or are there plans) to incorporate GPS priors earlier in the pipeline — for example, using them during initial camera pose estimation or optimization within PixSfM itself?
This would enable georeferenced reconstruction in a more direct and unified workflow.
Any suggestions, workarounds, or future features on this topic would be greatly appreciated!
Thanks again for your time and great project.