glomap using pose_prior

[starlordlxt]

When using pose prior with glomap, I input the world to camera poses and obtained very chaotic results.I first modified database.db with pose prior, then I used the command line:
glomap mapper --database_path dataset/orb_data/distorted/database1.db --image_path dataset/orb_data/inputs --output_path dataset/orb_data/distorted/sparse --GlobalPositioning.optimize_positions 0 --constraint_type ONLY_POINTS

I am very sure that my w2c is correct because I completed the sparse reconstruction using triangulation mode in colmap. Additionally, I switched my branch to /user/joschonb/pose-prior-fix.
colmap point_triangulator --database_path distorted/database.db --image_path inputs --input_path distorted/sparse/1 --output_path distorted/sparse/2

Tks! Looking forward to your advice!

[lpanaf]

Hi @starlordlxt, which version of GLOMAP are your running? The one in the main branch? If so, can you maybe checkout the one in the pull request? We put some fix there for this, and we would love to have your feedback

[ahojnnes]

@starlordlxt Please note that colmap now also has a prior-based incremental mapper (only a position prior) that could likely produce good results.

[mazy1998]

Hi team, it was really nice to meet y'all at ECCV. I was wondering when a tutorial or CLI would come out for GLOMAP given pose priors, im confident of my priors given that they can train a stable but blurry GS. However, I just want to refine them with the GLOMAP algorithm, currently with spatial mapper GLOMAP takes 3 hours for 4,000 images I think with rotation priors this can you down much more.

If I can do anything to help with this endeavor please let me know.

[jianghr-shanghaitech]

Wondering if this is solved, and how to use pose_prior collected by mobile phones' imu

[reynoldscem-oculo]

@ahojnnes Is there documentation for the prior-based mapper anywhere? I don't seem able to find anything on it.