Cars

Added

- DEM Generation with Pandora matches from low resolution images [943].
- Add global memory and cpu usage profiling.
- Beefrost integration (External Dem reprojection on sensor and epipolar geometry) [904].
- Phased DSMs re-entrance [967].
- Re-entrance of sensor images in pipelines starting on depth maps or DSMs [981].
- Classified performance map [966].
- Colorisation and classification from sensor images after DSM generation [938].

Changed

- Auto mode modified [1004].
- Performance map refactored [977].
- Point cloud outlier removal activated by default.
- C++ optimisations [925].
- Ambiguity can be added in output auxiliary data. [974]
- "Pipeline" and "geometry_plugin" in "advanced" configuration [990].
- Cubic interpolation used in epipolar grid interpolation.
- Refactoring of documentation [955].
- User can change multiprocessing mode [996].
- Sift matches are decimated [975].
- Inverse localisation for disparity grid, are parallelized [870].
- Removal of point cloud outlier filtering after point cloud fusion.
- Update Pandora version [918].
- Update Bulldozer version [1012].


Fixed

- Outlier removal bug fix.
- Worldwide vrt usage freeze fix [936].

Added

- Add cars-devibrate script (based on the work of Julien Michel)

Fixed

- Fix classification holes in epipolar geometry [949]

Fixed

- Fix wheel by using importlib instead of setuptools_scm
- Cast color before fusion to avoid asymetrical behavior [954]
- Adapt cars-starter to the new CARS interface
- Documentation corrections

Added

- Add `bulldozer-dtm` dependency to cars with a new bulldozer base DSM filling application [901]
- Add `cars-filter` dependency with outlier filtering functionalities in epipolar geometry [867]
- New parameter `use_cross_validation` in dense matching to manage Pandora cross validation [900]
- New parameters `altitude_delta_min` and `altitude_delta_max` in initial_elevation [902]

Changed

- Cars pipelines have been replaced by a "default" pipeline [895]
- Support for Python version superior than 3.10 with libsgm 0.5.0 [912]
- Support for Python 3.8 is over [930]
- The `no merging` mode is now the operational mode in cars, outlier filtering is done in epipolar geometry and laz are now saved in triangulation application [895]
- Add an index.json file in cars official product [893]
- Improve formatting of cars output products, improve cleaning of the output directory [874]
- Improve SIFT default configuration [928]
- Optimize SIFT filtering and grid correction [869]
- Optimize classification resampling [835]
- Simplify the check input function [920]
- Support `z_inf` and `z_sup` in depth map inputs [897]
- User and developer documentation updates [935] and [885]
- Use shareloc 0.2.4

Fixed

- Typo fixes (with API changes) [934]
- `dense_matches_filling` application renamed into `dense_match_filling`
- `holes_detection` application renamed into `hole_detection`
- `point_cloud_outliers_removing` renamed into `point_cloud_outlier_removal`
- `points_cloud` renamed into `point_cloud` in the code
- Fix crashes during report generation [916]
- Fix cars when using image with negative y axis [924]
- Fix no-data handling in elevation data [922]
- Make cars robust to different classification geotransforms [923]

Added

- Usage of up/down sampled sensor images [797]
- Pleiades usage tutorial [792]

Changed

- Refactoring of input and output configuration [824]
- Choose interpolation mode used in resampling [831]
- Displacement grids changed to localisation grids [828]
- Cars-resample and Cars-rasterize optimisation [803]
- Dsm radius parameter can now be used as a float [792]
- Refactoring of geoid usage [830]

Fixed

- Artefacts in performance map [846]
- Performance map modified [898]
- DSM clipping removed in notebook [853]
- Dask / pandas future warning removed [759]
- Crash avoided when temporary files lost [847]
- Blue pixels in orthorectified image [812]

Added

- Auto mode for orchestrator adapts to machine resources [798]

Changed

- Orfeo Toolbox geometry plugin is no longer available [805]
- Exogenous DEM is cropped before disparity grid generation to reduce time and memory consumption [827]

Fixed

- Subpixellic shift during resampling is fixed [799]
- Input mask can be saved as output of dense_point_clouds_to_dense_dsm pipeline [816]
- Ambiguity is normalized globally and no longer by tiles [810]
- Classified pixels outside of epipolar footprint are no longer filled [814]
- Statistics on confidence intervals are added and warnings are removed [817]
- Parameter check_inputs can be used with Shareloc geometry plugin [820]
- Masked and filled pixels are no longer masked in final DSM [836]
- Cache of GDAL is limited to 500Mb to avoid memory overload during resampling [823]