Support for dual-camera datasets (stereo/multi-camera reconstruction)

[DuVogel87]

Hello,

I would like to ask how to use dual-camera datasets with HLOC.
In my case, I have two cameras capturing images simultaneously (stereo setup).

With tools like COLMAP or Agisoft Metashape, handling such datasets works fine, but for HLOC I couldn’t find any documentation or examples regarding multi-camera support.

Question

• How can I reconstruct images captured from two cameras simultaneously with HLOC?
• Is there a recommended workflow for multi-camera systems (e.g., stereo rigs), or do I need to adapt the pipeline manually?

This is the pipeline I currently use (paths are kept generic here):

# Feature extraction
python -m hloc.extract_features \
    --conf superpoint_aachen \
    --image_dir path/to/images \
    --export_dir path/to/outputs/sfm/

# Image retrieval and pair selection
python -m hloc.extract_features \
    --conf netvlad \
    --image_dir path/to/images \
    --export_dir path/to/outputs/sfm/

python -m hloc.pairs_from_retrieval \
    --descriptors path/to/outputs/sfm/global-feats-netvlad.h5 \
    --output path/to/outputs/sfm/pairs-netvlad.txt \
    --num_matched 5

# Feature matching
python -m hloc.match_features \
    --conf superglue \
    --pairs path/to/outputs/sfm/pairs-netvlad.txt \
    --features feats-superpoint-n4096-r1024 \
    --export_dir path/to/outputs/sfm/

# SfM reconstruction
python -m hloc.reconstruction \
    --sfm_dir path/to/outputs/sfm/sfm_superpoint+superglue \
    --image_dir path/to/images \
    --pairs path/to/outputs/sfm/pairs-netvlad.txt \
    --features path/to/outputs/sfm/feats-superpoint-n4096-r1024.h5 \
    --matches path/to/outputs/sfm/feats-superpoint-n4096-r1024_matches-superglue_pairs-netvlad.h5

# Example for fisheye
python -m hloc.reconstruction \
    --sfm_dir path/to/outputs/sfm/sfm_superpoint+superglue \
    --image_dir path/to/images \
    --pairs path/to/outputs/sfm/pairs-netvlad.txt \
    --features path/to/outputs/sfm/feats-superpoint-n4096-r1024.h5 \
    --matches path/to/outputs/sfm/feats-superpoint-n4096-r1024_matches-superglue_pairs-netvlad.h5 \
    --camera_mode SINGLE \
    --image_options camera_model="'OPENCV_FISHEYE'"

[lukechencqu]

did you solved this problem?
the lastest colmap 3.13.0 support rig sfm. before that at feature extraction step, colmap bind each camera with a rig by default, while HLoc did not.
so, i think it should modify at the feature extraction step of Hloc(bind each camera with a rig) and write the info into DB.

[DuVogel87]

No.
But as you said, i have tried Colmap Rig support and it works for me.