.. _cli: Command-line Interface ====================== The command-line interface provides access to all of COLMAP's functionality for automated scripting. Each core functionality is implemented as a command to the ``colmap`` executable. Run ``colmap -h`` to list the available commands (or ``COLMAP.bat -h`` under Windows). Note that if you run COLMAP from the CMake build folder, the executable is located at ``./src/colmap/exe/colmap``. To start the graphical user interface, run ``colmap gui``. Example ------- Assuming you stored the images of your project in the following structure:: /path/to/project/... +── images │   +── image1.jpg │   +── image2.jpg │   +── ... │   +── imageN.jpg The command for the automatic reconstruction tool would be:: # The project folder must contain a folder "images" with all the images. $ DATASET_PATH=/path/to/project $ colmap automatic_reconstructor \ --workspace_path $DATASET_PATH \ --image_path $DATASET_PATH/images Note that any command lists all available options using the ``-h,--help`` command-line argument. In case you need more control over the individual parameters of the reconstruction process, you can execute the following sequence of commands as an alternative to the automatic reconstruction command:: # The project folder must contain a folder "images" with all the images. $ DATASET_PATH=/path/to/dataset $ colmap feature_extractor \ --database_path $DATASET_PATH/database.db \ --image_path $DATASET_PATH/images $ colmap exhaustive_matcher \ --database_path $DATASET_PATH/database.db $ mkdir -p $DATASET_PATH/sparse $ colmap mapper \ --database_path $DATASET_PATH/database.db \ --image_path $DATASET_PATH/images \ --output_path $DATASET_PATH/sparse $ mkdir -p $DATASET_PATH/dense $ colmap image_undistorter \ --image_path $DATASET_PATH/images \ --input_path $DATASET_PATH/sparse/0 \ --output_path $DATASET_PATH/dense \ --output_type COLMAP \ --max_image_size 2000 $ colmap patch_match_stereo \ --workspace_path $DATASET_PATH/dense \ --workspace_format COLMAP \ --PatchMatchStereo.geom_consistency true $ colmap stereo_fusion \ --workspace_path $DATASET_PATH/dense \ --workspace_format COLMAP \ --input_type geometric \ --output_path $DATASET_PATH/dense/fused.ply $ colmap poisson_mesher \ --input_path $DATASET_PATH/dense/fused.ply \ --output_path $DATASET_PATH/dense/meshed-poisson.ply $ colmap delaunay_mesher \ --input_path $DATASET_PATH/dense \ --output_path $DATASET_PATH/dense/meshed-delaunay.ply # Optionally simplify a dense mesh to reduce its size. $ colmap mesh_simplifier \ --input_path $DATASET_PATH/dense/meshed-poisson.ply \ --output_path $DATASET_PATH/dense/meshed-poisson-simplified.ply \ --MeshSimplification.target_face_ratio 0.25 # Optionally texture a mesh using the undistorted images. $ colmap mesh_texturer \ --workspace_path $DATASET_PATH/dense \ --input_path $DATASET_PATH/dense/meshed-poisson.ply \ --output_path $DATASET_PATH/dense/textured To use the global SfM pipeline instead of the incremental mapper, replace the ``mapper`` step with ``global_mapper``. The global mapper depends on good focal length priors, so if reliable intrinsics are not available (e.g., from EXIF or lab calibration), you should run ``view_graph_calibrator`` first. This step is optional but recommended to improve the quality of global SfM, as was always the default in `GLOMAP `_. Note that ``view_graph_calibrator`` modifies camera intrinsics and two-view geometries in the database in-place, so it is recommended to work on a copy of the database:: $ colmap feature_extractor \ --database_path $DATASET_PATH/database.db \ --image_path $DATASET_PATH/images $ colmap exhaustive_matcher \ --database_path $DATASET_PATH/database.db # Optional but often needed: calibrate intrinsics from the view graph. # This modifies the database in-place, so work on a copy. $ cp $DATASET_PATH/database.db $DATASET_PATH/database_global.db $ colmap view_graph_calibrator \ --database_path $DATASET_PATH/database_global.db $ mkdir -p $DATASET_PATH/sparse $ colmap global_mapper \ --database_path $DATASET_PATH/database_global.db \ --image_path $DATASET_PATH/images \ --output_path $DATASET_PATH/sparse If you want to run COLMAP on a computer without an attached display (e.g., cluster or cloud service), COLMAP automatically switches to use CUDA if supported by your system. If no CUDA enabled device is available, you can manually select to use CPU-based feature extraction and matching by setting the ``--FeatureExtraction.use_gpu 0`` and ``--FeatureMatching.use_gpu 0`` options. Help ---- The available commands can be listed using the command:: $ colmap help Usage: colmap [command] [options] Documentation: https://colmap.github.io/ Example usage: colmap help [ -h, --help ] colmap gui colmap gui -h [ --help ] colmap automatic_reconstructor -h [ --help ] colmap automatic_reconstructor --image_path IMAGES --workspace_path WORKSPACE colmap feature_extractor --image_path IMAGES --database_path DATABASE colmap exhaustive_matcher --database_path DATABASE colmap mapper --image_path IMAGES --database_path DATABASE --output_path MODEL ... Available commands: help gui automatic_reconstructor bundle_adjuster color_extractor database_cleaner database_creator database_merger delaunay_mesher exhaustive_matcher feature_extractor feature_importer geometric_verifier global_mapper guided_geometric_verifier hierarchical_mapper image_deleter image_filterer image_rectifier image_registrator image_undistorter image_undistorter_standalone mapper matches_importer mesh_simplifier mesh_texturer model_aligner model_analyzer model_clusterer model_comparer model_converter model_cropper model_merger model_orientation_aligner model_splitter model_transformer patch_match_stereo point_filtering point_triangulator pose_prior_mapper poisson_mesher project_generator rig_configurator rotation_averager sequential_matcher spatial_matcher stereo_fusion transitive_matcher view_graph_calibrator vocab_tree_builder vocab_tree_matcher vocab_tree_retriever And each command has a ``-h,--help`` command-line argument to show the usage and the available options, e.g.:: $ colmap feature_extractor -h Options can either be specified via command-line or by defining them in a .ini project file passed to ``--project_path``. -h [ --help ] --default_random_seed arg (=0) --log_target arg (=stderr_and_file) {stderr, stdout, file, stderr_and_file} --log_path arg --log_level arg (=0) --log_severity arg (=0) 0:INFO, 1:WARNING, 2:ERROR, 3:FATAL --log_color arg (=1) --project_path arg --database_path arg --image_path arg --camera_mode arg (=-1) --image_list_path arg --descriptor_normalization arg (=l1_root) {'l1_root', 'l2'} --ImageReader.mask_path arg --ImageReader.camera_model arg (=SIMPLE_RADIAL) --ImageReader.single_camera arg (=0) --ImageReader.single_camera_per_folder arg (=0) --ImageReader.single_camera_per_image arg (=0) --ImageReader.existing_camera_id arg (=-1) --ImageReader.camera_params arg --ImageReader.default_focal_length_factor arg (=1.2) --ImageReader.camera_mask_path arg --FeatureExtraction.type arg (=SIFT) --FeatureExtraction.max_image_size arg (=3200) --FeatureExtraction.num_threads arg (=-1) --FeatureExtraction.use_gpu arg (=1) --FeatureExtraction.gpu_index arg (=-1) --SiftExtraction.max_num_features arg (=8192) --SiftExtraction.first_octave arg (=-1) --SiftExtraction.num_octaves arg (=4) --SiftExtraction.octave_resolution arg (=3) --SiftExtraction.peak_threshold arg (=0.0066666666666666671) --SiftExtraction.edge_threshold arg (=10) --SiftExtraction.estimate_affine_shape arg (=0) --SiftExtraction.max_num_orientations arg (=2) --SiftExtraction.upright arg (=0) --SiftExtraction.domain_size_pooling arg (=0) --SiftExtraction.dsp_min_scale arg (=0.16666666666666666) --SiftExtraction.dsp_max_scale arg (=3) --SiftExtraction.dsp_num_scales arg (=10) The available options can either be provided directly from the command-line or through a ``.ini`` file provided to ``--project_path``. Commands -------- The following list briefly documents the functionality of each command, that is available as ``colmap [command]``: - ``gui``: The graphical user interface, see :ref:`Graphical User Interface ` for more information. - ``automatic_reconstructor``: Automatically reconstruct sparse and dense model for a set of input images. Key options include ``--quality`` (LOW, MEDIUM, HIGH, EXTREME), ``--data_type`` (INDIVIDUAL, VIDEO, INTERNET) to tune settings for different capture scenarios, ``--feature`` (SIFT, ALIKED) to select the feature extraction algorithm, ``--mapper`` (INCREMENTAL, HIERARCHICAL, GLOBAL) to choose the SfM pipeline, and ``--mesher`` (POISSON, DELAUNAY) to select the surface reconstruction method. - ``project_generator``: Generate project files at different quality settings. - ``feature_extractor``, ``feature_importer``: Perform feature extraction or import features for a set of images. - ``exhaustive_matcher``, ``vocab_tree_matcher``, ``sequential_matcher``, ``spatial_matcher``, ``transitive_matcher``, ``matches_importer``: Perform feature matching after performing feature extraction. - ``geometric_verifier``: Run standalone geometric verification on existing feature matches in the database. This estimates two-view geometries (fundamental/essential matrices, homographies) for matched image pairs. - ``guided_geometric_verifier``: Run geometric verification guided by an existing sparse reconstruction. Uses the known relative camera poses to improve match verification results. - ``mapper``: Sparse 3D reconstruction / mapping of the dataset using SfM after performing feature extraction and matching. - ``global_mapper``: Sparse 3D reconstruction using the global SfM pipeline. Unlike the incremental ``mapper``, the global approach solves for all camera poses simultaneously using rotation averaging and global positioning. This can be faster for large datasets but may be less robust to outliers. The global mapper depends on reasonably good focal length priors to perform well. Run ``view_graph_calibrator`` before ``global_mapper`` to calibrate camera intrinsics and estimate relative poses from the view graph, or provide camera calibrations manually. - ``pose_prior_mapper``: Sparse 3D reconstruction / mapping using pose priors. - ``hierarchical_mapper``: Sparse 3D reconstruction / mapping of the dataset using hierarchical SfM after performing feature extraction and matching. This parallelizes the reconstruction process by partitioning the scene into overlapping submodels and then reconstructing each submodel independently. Finally, the overlapping submodels are merged into a single reconstruction. It is recommended to run a few rounds of point triangulation and bundle adjustment after this step. - ``image_undistorter``: Undistort images and/or export them for MVS or to external dense reconstruction software, such as CMVS/PMVS. - ``image_rectifier``: Stereo rectify cameras and undistort images for stereo disparity estimation. - ``image_filterer``: Filter images from a sparse reconstruction. - ``image_deleter``: Delete specific images from a sparse reconstruction. - ``patch_match_stereo``: Dense 3D reconstruction / mapping using MVS after running the ``image_undistorter`` to initialize the workspace. - ``stereo_fusion``: Fusion of ``patch_match_stereo`` results into to a colored point cloud. - ``poisson_mesher``: Meshing of the fused point cloud using Poisson surface reconstruction. - ``delaunay_mesher``: Meshing of the reconstructed sparse or dense point cloud using a graph cut on the Delaunay triangulation and visibility voting. - ``mesh_simplifier``: Simplify a triangle mesh (PLY format) using Quadric Error Metric (QEM) decimation. This reduces the number of faces in a mesh while preserving its overall shape and appearance. Key options include ``--MeshSimplification.target_face_ratio`` to control the fraction of faces to retain (default 0.1), ``--MeshSimplification.max_error`` to set a maximum quadric error threshold (0 = disabled), and ``--MeshSimplification.boundary_weight`` to control boundary edge preservation (default 1000). Supports multi-threaded initialization via ``--MeshSimplification.num_threads``. - ``mesh_texturer``: Produce a texture atlas and UV coordinates for a triangle mesh using calibrated multi-view images. - ``image_registrator``: Register new images in the database against an existing model, e.g., when extracting features and matching newly added images in a database after running ``mapper``. Note that no bundle adjustment or triangulation is performed. - ``point_triangulator``: Triangulate all observations of registered images in an existing model using the feature matches in a database. - ``point_filtering``: Filter sparse points in model by enforcing criteria, such as minimum track length, maximum reprojection error, etc. - ``bundle_adjuster``: Run global bundle adjustment on a reconstructed scene, e.g., when a refinement of the intrinsics is needed or after running the ``image_registrator``. - ``database_cleaner``: Clean specific or all database tables. - ``database_creator``: Create an empty COLMAP SQLite database with the necessary database schema information. - ``database_merger``: Merge two databases into a new database. Note that the cameras will not be merged and that the unique camera and image identifiers might change during the merging process. - ``model_analyzer``: Print statistics about reconstructions. - ``model_clusterer``: Split a reconstruction into smaller sub-model clusters. Useful for managing and processing large-scale reconstructions. - ``model_aligner``: Align/geo-register model to coordinate system of given camera centers. - ``model_orientation_aligner``: Align the coordinate axis of a model using a Manhattan world assumption. - ``model_comparer``: Compare statistics of two reconstructions. - ``model_converter``: Convert the COLMAP export format to another format, such as PLY or NVM. - ``model_cropper``: Crop model to specific bounding box described in GPS or model coordinate system. - ``model_merger``: Attempt to merge two disconnected reconstructions, if they have common registered images. - ``model_splitter``: Divide model in rectangular sub-models specified from file containing bounding box coordinates, or max extent of sub-model, or number of subdivisions in each dimension. - ``model_transformer``: Transform coordinate frame of a model. - ``color_extractor``: Extract mean colors for all 3D points of a model. - ``rig_configurator``: Configure rigs and frames after feature extraction. - ``vocab_tree_builder``: Create a vocabulary tree from a database with extracted images. This is an offline procedure and can be run once, while the same vocabulary tree can be reused for other datasets. Note that, as a rule of thumb, you should use at least 10-100 times more features than visual words. Pre-trained trees can be downloaded from https://demuc.de/colmap/. This is useful if you want to build a custom tree with a different trade-off in terms of precision/recall vs. speed. - ``vocab_tree_retriever``: Perform vocabulary tree based image retrieval. - ``rotation_averager``: Run standalone rotation averaging on the view graph. Estimates global camera rotations from pairwise relative rotations. - ``view_graph_calibrator``: Calibrate camera intrinsics using the view graph. Estimates focal lengths and other intrinsic parameters from pairwise geometric relations. Should be run before ``global_mapper``, if no good prior camera intrinsics are known, since the global mapper depends on reasonably good focal length priors to perform well. Visualization ------------- If you want to quickly visualize the outputs of the sparse or dense reconstruction pipelines, COLMAP offers you the following possibilities: - The sparse point cloud obtained with the ``mapper`` can be visualized via the COLMAP GUI by importing the model files: choose ``File > Import Model`` and select the folder containing the sparse model files (``cameras.txt``, ``images.txt``, ``points3D.txt``, etc.). - The dense point cloud obtained with the ``stereo_fusion`` can be visualized via the COLMAP GUI by importing ``fused.ply``: choose ``File > Import Model from...`` and then select the file ``fused.ply``. - The dense mesh model ``meshed-*.ply`` obtained with the ``poisson_mesher`` or the ``delaunay_mesher`` can currently not be visualized with COLMAP, instead you can use an external viewer, such as Meshlab. Use the ``mesh_simplifier`` command to reduce the mesh size for faster visualization or downstream processing. Use the ``mesh_texturer`` command to produce a textured mesh with a texture atlas that can be visualized in Meshlab or other 3D viewers.