mrcal 2.4 release notes

New in mrcal 2.4

This is yet another mostly-maintenance release. A number of bugs are fixed, and a number of new features have been added. The biggest new features:

mrcal can be built with clang. Try it out like this: CC=clang CXX=clang++ make. This opens up some portability improvements, such as making it easier to run on Windows.
Full dense stereo pipeline in C.

All the tools use /usr/bin/env in the #! line. People who live life on hard mode now have it slightly easier
Tools to support more file formats:
These are experimental. Please let me know if these are or aren't useful
mrcal-stereo, mrcal.stereo_range() work properly if disparity_min>0
The mrcal-stereo GUI displays the pixel coordinates for both cameras, not just the one under the cursor

The Procrustes fit functions (mrcal.align_procrustes_vectors_R01(), mrcal.align_procrustes_vectors_Rt01()) detect and report errors in the input data
mrcal.show_projection_diff() has similar contour-generating options as what mrcal.show_projection_uncertainty() has
renamed mrcal.residuals_chessboard() to mrcal.residuals_board(). Both names are available for backwards compatibility
mrcal.show_residuals_...() functions and mrcal-show-residuals-... tools now all have cbmax and --cbmax options respectively to set the upper bound for the mapping in the heat map display
Added mrcal.CHOLMOD_factorization.rcond() as an interface to cholmod_rcond()
mrcal.worst_direction_stdev() works with NxN arrays, not just 2x2
mrcal.ref_calibration_object() has a new optimization_inputs argument. This provides a convenient shorthand to get the object used in a particular calibration
Added mrcal.R_aligned_to_vector() to produce a non-unique rotation to map a given vector to [0,0,1]
Added mrcal.sorted_eig() to compute an eigenvalue decomposition with sorted eigenvalues
Added mrcal.write_point_cloud_as_ply() to write a point cloud as a .ply file
mrcal.apply_color_map() supports all gnuplot rgbformulae, above the 7,5,15 default
mrcal.stereo_range() has more optional arguments:
- disparity_max to set the upper valid-disparity bound
- disparity_scaled_min, disparity_scaled_max to act on the raw disparity data produced by the stereo matcher

Better error handling: lensmodel-parsing functions return an error code instead of asserting, if given invalid input. Affected functions:
- mrcal_lensmodel_name_unconfigured()
- mrcal_lensmodel_name()
- mrcal_lensmodel_type_from_name()
- mrcal_lensmodel_from_name()
More image types supported. We now have:
- mrcal_image_int8_t
- mrcal_image_uint8_t
- mrcal_image_int16_t
- mrcal_image_uint16_t
- mrcal_image_int32_t
- mrcal_image_uint32_t
- mrcal_image_int64_t
- mrcal_image_uint64_t
- mrcal_image_float_t
- mrcal_image_double_t
- mrcal_image_bgr_t
Renamed the bgr_t type to mrcal_bgr_t. For namespacing.
mrcal_rectification_maps() supports LENSMODEL_LATLON and LENSMODEL_PINHOLE rectification
The remaining pieces for a full dense-stereo pipeline in C were implemented. The new functions are
- mrcal_stereo_range_sparse() to convert a set of disparity values to ranges
- mrcal_stereo_range_dense() to convert a disparity image to a range image
- mrcal_apply_color_map() to convert a disparity or range image to a heat map for human consumption
A dense-stereo-in-C sample is provided.

C API/ABI breaking change: these functions now take an extra int Nobservations_board argument:
- mrcal_pack_solver_state_vector()
- mrcal_unpack_solver_state_vector()
C API breaking change: bgr_t structure renamed to mrcal_bgr_t
Python API: a function was renamed: mrcal.residuals_chessboard() -> mrcal.residuals_board(). Both names are available for backwards compatibility, but moving to the new name is recommended.
Python mrcal.apply_color_map() function: a_min, a_max arguments now must be passed in as kwargs, and not as positionals.