Refinement of unoptimized camera poses

[bkhanal-11]

Thank you for this amazing work.

I have images, poses and intrinsic matrix obtained from a VIO SLAM. The poses I have are not optimized. I was looking for a tools to refine those poses based on feature matching from those images. Could you help me guide through it? I previously tried to use COLMAP and its instruction to refine pose but got really poor result.

So far I have tried to extract and match features using COLMAP, parse the known poses, images and intrinsics in the format "cameras.txt, images.txt, points3d.txt", then used point_triangulator and finally pixsfm.refine_colmap bundle_adjuster. But the results are worst than initial poses.

[Yuhuoo]

I also have the same question.