Closed oneleggedredcow closed 3 months ago
Yes: incremental_mapping
subsequently refines the points and camera poses with bundle adjustment: https://github.com/colmap/colmap/blob/06a805386afb3db9dc6704c7ba2a80861bb64f7e/src/colmap/sfm/incremental_mapper.cc#L572-L577
You can accomplish this in Python using pyceres, see the BA example (section Optimize 3D points). Alternatively we could bind colmap/estimators/bundle_adjustment.h
.
Awesome, thank you!
I'm trying to use estimate_triangulation to triangulate a set of points. To test that I understood how to use it correctly, I took the results from incremental_mapping and used that as input to estimate_triangulation. Unfortunately, I'm unable to obtain the same result from estimate_triangulation as I get from incremental_mapping.
The reprojection error from the estimate_triangulation point is slightly higher than the error from incremental_mapping point. Why do they not produce the same result?
I thought it might have to do with the camera_model so I switched all of the camera models to SIMPLE_PINHOLE and I get the same result.
I tried triangulating the points using a different approach (https://gist.github.com/davegreenwood/e1d2227d08e24cc4e353d95d0c18c914) and that gives me a very similar answer to estimate_triangulation. Is there something extra that I need to do in order to get the more accurate result that incremental_mapping provides?