learning-photogrammetry

This repository contains practical examples of how to use colmap, an open-source general general-purpose Structure-from-Motion (SfM) and Multi-View Stereo (MVS) pipeline with docker.

Extracting features

Attention to the GPU flag. If you have a GPU, set it to 1, otherwise set it to 0.

docker run --init --rm --gpus all -v .:/data colmap/colmap colmap feature_extractor --database_path /data/database.db --image_path /data/imgs/ --FeatureExtraction.use_gpu 1

Matching features

docker run --init --rm --gpus all -v .:/data colmap/colmap colmap exhaustive_matcher --database_path /data/database.db --FeatureMatching.use_gpu 1

docker run --init --rm --gpus all -v .:/data colmap/colmap colmap sequential_matcher --database_path /data/database.db --FeatureMatching.use_gpu 1

Sparse reconstruction

mkdir -p sparse
docker run --init --rm --gpus all -v .:/data colmap/colmap colmap mapper --database_path /data/database.db --image_path /data/imgs/ --output_path /data/sparse/

Sparse PLY conversion (Not needed for further steps)

docker run --init --rm --gpus all -v .:/data colmap/colmap colmap model_converter --input_path /data/sparse/0 --output_path /data/sparse/0/sparse.ply --output_type PLY

Image undistortion

mkdir -p dense
docker run --init --rm --gpus all -v .:/data colmap/colmap colmap image_undistorter --image_path /data/imgs --input_path /data/sparse/0 --output_path /data/dense --output_type COLMAP --max_image_size 2000

Dense reconstruction (+GPU)

docker run --init --rm --gpus all -v .:/data colmap/colmap colmap patch_match_stereo --workspace_path /data/dense --workspace_format COLMAP --PatchMatchStereo.geom_consistency true

docker run --init --rm --gpus all -v .:/data colmap/colmap colmap stereo_fusion --workspace_path /data/dense --workspace_format COLMAP --input_type geometric --output_path /data/dense/fused.ply

Visualization

uv run open_ply.py --ply <path_to_ply>

Optional point-cloud subsampling:

uv run open_ply.py --ply <path_to_ply> --voxel-size 0.02