0.17 release

• Added compatibility with cfitsio syntax for selecting a hdu in detection images
• Added support for multi-extension fits detection images, to be processed in sequence and get catalog with parts
• Added ErrorEllipse property
• Added model fitting without psf, use "nospsf" as filename to activate
• Stability improvements

0.16 release

New features:

• Model fitting improvements
  • Iterative deblending for model fitting
  • Custom user-made models can be provided in the form of ONNX compute graphs
  • Automatic fit resolution downgrade for very large models
  • Python priors are now evaluated in the C++ code for increased performance
  • Overall major improvements to stability, performance and memory usage

Changes that may require configuration files update:

• Multi-thresholding and grouping of split sources are now on by default
• In the model fitting Python configuration, source properties are now actual properties instead of getter functions: o.get_radius() => o.radius
• Iterative model fitting is on by default, if needed the new system can be turned off with use_iterative_fitting(False) (not all improvements from this release are available in the old system)
• Segmentation using a ML model was renamed from --segmentation-onnx-model to --segmentation-ml-model (experimental feature)
• The Onnx property that performs a measurement using an Onnx compute graph is renamed to MLMeasurement

0.15 release

Compatibility breaking change since 0.14:

FITS HDU numbering scheme in the python configuration file has been changed to start at 0 for the primary HDU instead of 1.

While 1 based indices are used internally by the cfitsio library we realized that 0 based indices are much more common. In case of manually specified HDU numbers, some Python configuration files may need to be adjusted.

New features:

• PSF stacks
• Association mode

Other changes:
Bug fixes and performance improvements especially on machines with lots of CPU cores
SourceXtractor++ now requires version 2.19 of the Alexandria library.

0.14 release

This release fixes a performance regression in 0.13 as well as some missing error bars in aperture photometry

0.13 Release

Improvements:

• Multi-threaded computation for grouping/cleaning
• Variable PSF: Add support for other coordinates
• New segmentation mode: BFS, uses a combination of breadth first search and tile by tile processing using a Hilbert curve to optimize memory usage

Fixes:

• Python configuration now uses internal methods instead of astropy to load FITS files to avoid a few issues caused by the difference
• Fixed a problem with multiple dots in checkimage filenames
• Fixed a regression related to maximum number of simultaneous open files
• Fixed a problem with zero values in weight images
• Moffat checkimage parameter renamed to --debug-image-moffat to reflect its meant for internal testing use (very slow performance)
• Unsigned int check images are now saved as signed int check images to avoid compatibility issues
• Improved WCS error reporting

0.12 Release

• Added model fitting re-normalization to fix accuracy issues and inconsistent error bars
• New option: model fitting resolution multiplier (--sampling-scale-factor)
• Better default parameters for model fitting sampling
• ONNX plugin, can load ONNX format neural networks and perform inference on all sources
• Fixed build on Fedora 33
• output format option is no longer case sensitive

0.11 release

Bug fixes

• Fix handling of HDU number when using MEF files (#260)
• Add support for comment on .head files (#242 and #260)
• Fix for wcslib 7 (#232)
• Fix for background cells with 0 sigma (#233 and #234)
• Fix infinities on auto_flux_err when the gain is 0 (#239)
• Improved performance when reading the header from MEF files (#237)
• Fix handling of cell sizes for the background model (#245, #247)
• Use the fftw3 mode FFTW_ESTIMATE instead of FFTW_MEASURE. Solve reproducibility issues. (#256)
• Catch errors from wcslib (#259)
• Fix computation of uncertainties in the model fitting (#263)
• Fix potential segfaults when running multithreaded (#265)
• Fix the format of the units written into the catalog (#268)

New Features

• Allow .head files to override WCS coordinates (#240)
• Generate elongation and ellipticity columns when the property ShapeParameters is enabled (#250)
• New plugin for Signal to Noise measurement: SNRRatio. Outputs the column snrratio (#251)
• New plugin Vignet. Generates a multidimensional column vignet: (n.images, width, height) (#252)
• New plugin GrowthCurve: Generates columns flux_growth and flux_growth_step. Growth curve is measured in each measurement frame (#264)
• New plugin FluxRadius. Generates a column flux_radius. Measured in each measurement frame. (#264)
• Added support for generating LDAC catalogs. It can be enabled with output-catalog-format=FITS_LDAC (#258)

Other

• Update to Alexandria 2.16
• Refactored background code (#248). If it gives trouble, it can be disabled with the flag --background-legacy
• Can build with C++14

0.10 release

Changes since 0.8:

• Incompatible change: the sign convention for the model fitting angle is now reversed. It now uses the same counter-clockwise convention as ShapeParameters and SExtractor 2. This change does not impact world coordinate results.
• Improved performance overall and especially for model-fitting
• Added /etc to the config search path
• prevent race condition with some versions of wcslib that could result in crashes and corrupted results in multi-threaded model-fitting
• fixed a problem with loading compressed multi-extension FITS files

0.9 test pre-release

(Note: odd-numbered versions are test pre-releases. The current stable version is 0.8. The next official release will be 0.10)

Changes since 0.8:

• Incompatible change: the sign convention for the model fitting angle is now reversed. It now uses the same counter-clockwise convention as ShapeParameters and SExtractor 2. This change does not impact world coordinate results.
• Improved performance overall and especially for model-fitting
• Added /etc to the config search path
• prevent race condition with some versions of wcslib that could result in crashes and corrupted results in multi-threaded model-fitting
• fixed a problem with loading compressed multi-extension FITS files

0.8 release

Changes since 0.7:

• Fixed isophotal variance when using gain=0 to indicate infinite gain

Changes since 0.6.1:

New features:

• Support for .head text files to override fits file headers
• New option --dump-default-config to produce a default configuration file
• Added --grouping-moffat-max-distance option to constrain Moffat grouping to a maximum distance in pixels
• Added an optional sliding window for Lutz segmentation (--segmentation-lutz-window-size), this allows measurement to start on sources that fall outside the window to save on memory and enable multithreading sooner. Off by default.
• Support for the GSL minimization engine as an alternative to levmar. Levmar is still the default and recommended.
• New signal to noise ratio check image (--check-image-snr)
• New core threshold cleaning method (--partition-corethreshold)

Bug fixes and minor changes:

• Fixed crash when Moffat model fitting starts with a negative value as flux guess
• Added grouping options into the help display
• Removed "sextractor" name from the code
• Fixed a crash when exiting if using external plugins
• Improved the robustness of multi-extension FITS loading
• Improved performance of Moffat model evaluation
• Will look for plugins in default directories too
• Fixed pixels with negative variance values sometimes appearing
• Compressed FITS files are now loaded correctly
• Improved error reporting from Python
• Outputs more and better background modelling statistics
• Fixed compilation issues on OSX
• Better multithreading performance if levmar was compiled with the proper options to be fully reentrant
• Made the code compatible with the Intel compiler
• Fixed default configuration file not found warnings
• Source ids now start at 1 instead of 0 (source 0 was previously not distinguishable from the background on the check images)
• The output catalog is now written incrementally instead of all at once (controllable with --output-flush-size)