v0.6.1

• Overload Camera.img_from_cam for 3D points (#256)
• Add constructor Rotation3d(angle_axis) and methods Rotation3d.{angle,angle_to} (#258)
• Bugfix: properties defined with def_readwrite now return writeable arrays (#257)

Full Changelog: v0.6.0...v0.6.1

v0.6.0

This release requires COLMAP 3.9.1.

• Make glog usable in Python (#237)
• Add example.py (#237)
• Add incremental mapping callbacks (#238)
• Minimal bindings for the database (#238)
• Bugfix in CreateSummary (#241)
• Make objects pickleable (#243)
  • To enable this, todict() now only exports data attributes
  • Rotation3d.quat becomes a property
  • {Rigid3d,Sim3d}.matrix becomes a function
  • Image.num_points2D becomes a function
• Bind essential matrix utilities (#244)
• Autocast numpy.ndarray <-> std::vector<Eigen::Vector> (#248)
• Transform many points at once (#250)
  • Remove Image.{project,transform_to_image,transform_to_world}

Full Changelog: v0.5.0...v0.6.0

v0.5.0

This release requires COLMAP 3.9.1.

• Propagate changes from COLMAP
  • New Rigid3d/Sim3d objects
  • CamFromWorld replaces qvec/tvec (#166)
  • WorldToImage/ImageToWorld become ImgFromCam/CamFromImg (#167)
  • Decouple SIFT matching from two view geometry options (#169)
• Improve the bindings
  • Rename alignment estimators to align_reconstructions_via_reprojections, align_reconstructions_via_proj_centers, and align_reconstructions_via_points (#174)
  • Estimators return None on failure (#175)
  • Better logging with glog (#176)
  • Add robust triangulation (#184)
  • Add bundle adjustment (#146)
  • Expose Incremental{Mapper,Triangulator}::Options (#219)
  • Better __repr__ and docstrings for many objects (#224)
  • Many bug fixes
• Update the file structure to follow the one of COLMAP
• Wheels:
  • Deprecate Python 3.6 and support Python 3.10
  • Support Windows (#214) and macOS 10-12 with Apple Silicon (#218)

v0.4.0

• Support for Ceres v2.2 (#87)
• Wheels: deprecate Python 3.6 and add Python 3.10 (#109)
• Add input model support in incremental mapping pipeline (#84)
• Add Poisson and Delaunay meshing (#93)
• Many bug fixes

v0.3.0

• import_images and infer_camera_from_image support ImageReaderOptions, for example to control the default camera (#74)
• Minor fixes to the macOS CI pipeline (#78)

v0.2.0

• Release the GIL in estimators for multi-threading (#40)
• Support building with CUDA (#57)
• More bindings:
  • CorrespondenceGraph and IncrementalTriangulators (#41)
  • IncrementalMapperOptions (#56)
  • Dense MVS reconstruction (undistort_images, patch_match_stereo, stereo_fusion) (#57)
  • SIFT feature extraction and matching (exhaustive, sequential, spatial, vocabtree) (#57)
  • GPU SIFT with unified interface (#57)
• Allow interrupting long C++ functions via KeyboardInterrupt (#57)
• Improve the wheels: add more Python versions, fix broken dependencies (#58, #51)

v0.1.0

• Add bindings for:
  • more COLMAP objects: Reconstruction, Image, Camera, Track, TrackElement, Point2D, Point3D, SimilarityTransform
  • steps of the reconstruction pipeline: image importer, two-view verification, point triangulation, incremental mapping
  • functions to project points, align reconstruction, infer camera parameters from EXIF metadata
  • options for all estimators
• Estimators accept a Camera or a dictionary
• Estimators accept an Option dataclass object or a dictionary
• Checks raise exceptions instead of terminating the Python interpreter/kernel
• COLMAP outputs can be redirected to Python stdout
• Improve the documentation

Most of these improvements were contributed by Philipp Lindenberger - thank you!

See PyPI for the compiled wheels: https://pypi.org/project/pycolmap/

v0.0.3

First official release. Continuous integration with automated build and push to Pypi.