Change-log of versions¶
0.21.0 20/01/2022¶
- One year of development: 523 commits, close to 100 pull-requests. +53000 lines of code and 20000 suppressed.
- Sigma-clipping allows separation of Bragg/amorphous signal:
- Implementation in Python, Cython and OpenCL with poissonian and azimuthal error-model
- Sparsification, compression of single crystal data
- Application to peak-picking and quality assessement of SSX data
- Analysis of grid-scan to find single crystal
- Single pass variance propagation in azimuthal bin
- Integration of the Jungfrau detector (ID29)
- 2D integration:
- New integrators with error propagation by default
- Full pixel splitting in addition to BBox and no splitting
- Refactor of all LUT and CSR to share the same code base, makes tests more robust.
- Calibration of experimental setup using Jupyter notebooks
- User interaction with plots in matplotlib (thanks Phil Hans)
- Factorize code between pyFAI-calib and jupyter calibration
- Tutorial as notebook and video recording
- Parallax correction for thick detector (still experimental, thanks to Vadim)
- Improved tutorial on detector geometry calibration (based on Kabsch alignment)
- Better performances on HPC nodes by limiting simple OpenMP to fewer cores
- Many improvement in test, typos fixed, doc …
- Deprecate all legacy integrators since the new version sees its matrix complete.
- Supports Python 3.6 … 3.10 under Windows, MacOS and Linux.
0.20.0 22/01/2021¶
- One year of development, about 500 commits & 370 files modified.
- Generalization of the new generation 1D integrators (better error propagation)
- Sigma clipping and sparsification of single crystal data (OpenCL only)
- Fix issue introduced with the scipy 1.15 (constrained calibration broken)
- Improved distortion correction (also on GPU, …)
- Major re-work of the documentation (thanks Thomas Kluyver and Loic Huder)
- Improve the calibration of Pilatus and Eiger detectors based on a grid of holes.
- New cylindrical detector from Rigaku
- Drop deprecated OpenCL integrator
- Support all Eiger2 detectors (thanks Clemens Weninger and Marie Ruat)
- CI: move to Gihub workflow and gitlab-runners (bob) for building (thanks Thomas Vincent).
- Build for debian 10 and 11 (also ubuntu 20.04), drop debian 9
- Remove Python2 related code
- Lower Numpy ABI dependency as much as possible (remains _distortion’s C++ code)
- Drop tests for Python 3.5, checked on 3.9 as well.
0.19.0 31/01/2020¶
- Minor revison with only 150 commits, mainly bug-fixes
- Improvement on the GUI with many small bug-fixes
- Support newer h5py (mode mandatory, [()], …)
- Build for debian 10 and 11 (also ubuntu 20.04)
- Drop tests for Python 2 and 3.4, checked on 3.8
- Improved compatibility with ImageD11
- Use hdf5plugin to provide hdf5 io-filters in apps
- Rework diffraction mapping tools to use a worker
- New generation azimuthal integrator using CSR algorithm implemented in Python, Cython and OpenCL.
- Sigma-clipping implemented in OpenCL
- Publication on new generation integration, the GUI for calibration and the goniometers accepted in J. Synch. Rad. DOI: 10.1107/S1600577520000776
- A big thank to Florian from Germany, Bertrand from Xenocs, Alex from Soleil and Jon from ESRF for their contributions.
0.18.0 15/05/2019¶
- Last release with Valentin as he finishes his contract soon
- almost 800 commits, 60 PR since the last release: this is a huge release !
- Major rework on all GUIs, mainly pyFAI-calib2 and pyFAI-integrate.
- Possibility to integrate image stacks (i.e. from HDF5), …
- Rework the method to specify the algorithm, pixel splitting and implementation with sensible fall-backs. Also available via the different GUIs
- 3D visualization of detectors and experimental setup, useful for non flat detectors.
- integrate1d_ng is available with histogramming without pixel splitting in Python, Cython and OpenCL. Now, propagates the variance properly !
- IO sub-packages with associated refactoring for ponifile, json, …
- Improved management of OpenMP: simplify the code for histograms.
- Improved geometry description and tutorial for writing exchange with other software (ImageD11, thanks to Carsten Detlefs).
- More reliable simple ellipse fitting with tests and doc.
- Better POCL integration (debugged on cuda, x87, Power9, …)
- Rely on silx mechanics for the build, test, download, GUI, opencl …
- Many new tutorials, now available on binder-hub: new calibrants, Pilatus calibration, …
- Fix many issues reported in third-party software (Dioptas, …)
- Drop support of debian8, Python 2.7 and Python 3.4 on all platforms. It is likely most functionalities still work but without guaranty.
0.17.0 19/12/2018¶
- Only 200 commits in a couple of month, this is a small release
- Fix major bugs in pyFAI-calib2 (double validator, initial guess, ring position)
- Constrains have been added to the geometry fitting of pyFAI-calib2
- New pyFAI-integrate graphical application
- Much better support for user defined detector (HDF5)
- Start the rewrite of all integrators to allow proper error propagation (2D done)
- Factorize the preprocessing steps for many integrators
- Remove tons of code which has been deprecated for years in AzimuthalIntegrator
- Featuring contribution from Soleil and Berkeley
- Stop supporting Python2.7 on Windows (there won’t be wheels!)
- All scripts are now using Python 3.x (x>=4)
- This is the last release supporting Python 2.7, 3.4 hence debian 8
0.16.0: 26/10/2018¶
- Almost 800 commits since 0.15 !
- Huge improvements on the graphical application for calibration
- New detector definition (with manufacturers)
- Improved tests: the GUI is now tested
- Preparation for changing all rebinning engines (see variance tutorials)
- Azimuthal integrators (and most other objects) are now serializable with Pickle
- New distortion correction using the SparseBuilder C++ code
- New PONI-file format (detector definition changed)
- Isocontour is now provided by silx
- Peak-picker clean up (better peak selection near gaps)
- new Goniometer refinement with enhanced rotation using Euler angles
- Updated documentation: new cookbooks and tutorials about: - The use of the calibration application (cookbook) - log-scale integration of SAXS data (notebook) - Variance propagation (notebook)
0.15.0: 01/02/2018¶
- 150 commits since last revision
- New tutorials on image inpainting, sensor thickness correction, …
- Improve scripts
- Improve the new calibration GUI (pyFAI-calib2)
- Use scipy physical constants instead of hard-coded values
- Improved detector serialization and binning assessement
- Update documentation (images, better rendering of notebook & tutorials)
- Converge project with silx and fabio
- Remove generated rst- and C-files from repository
- This is the last version supporting python2.7
0.14.2: 14/09/2017¶
- Fix seg-fault with manylinux1 wheels, in fastcrc module (thanks Thomas)
- Fix Qt4-Qt5 compatibility (thanks Vadim)
- Easier to understand geometry transformation (thanks Jon)
- Lower memory consumption, better cache management
- Unified debian packaging working for 6->9
- New detector: Mythen & CirPad (thanks Fred)
- Clean up debug code which avoid to use pyFAI-calib2
- pyFAI-calib2 now expect fabio >= 0.5
- Fix issue with metadata saving in 1d
- Fix performance regression with pyopencl >2015.2 (Thanks Andreas)
- pyFAI saxs and waxs scripts guess now the binning of the detector (thanks Fred).
0.14.1: 25/07/2017¶
- Fixes Debian 7 and 8 packages
0.14.0: 20/07/2017¶
- Graphical user interface for calibration (pyFAI-calib2)
- Goniometer calibration tools and multi-geometry enhancements
- Integration scripts and averaging scripts are now able to normalize the data from monitors found in the header.
- Propagate metadata information as part of the integrated data
- Common pre-processing factorization on Python, Cython and OpenCL
- Test clean up and acceleration (avoid tests on too large images)
- Many new tutorials http://pyfai.readthedocs.io/en/latest/usage/tutorial/index.html
- New averaging / integration methods: - Azimuthal median filtering - Azimuthal trimmed mean - sigma-clipping on azimuthal angle - Radial averaging
- Diffraction image inpainting to fill-up the gaps with plausible values.
- This release correspond to 572 commits
- Change of license: now all pyFAI is MIT license.
0.13.0: 01/12/2016¶
- Global improvement of tests, packaging, code quality, documentation and project tools
- Scripts
- Add support for multiframe formats on pyFAI-average
- Add support for monitoring correction from header file (on pyFAI-average)
- Add progressbar in the shell (on pyFAI-average and pyFAI-integrate)
- Script drawMask_pymca is renamed into pyFAI-drawmask
- Rework of the drawmask GUI using silx
- pyFAI-drawmask do not have anymore hard dependency on PyMCA
- pyFAI-integrate can now be used without qt dependency (–no-gui)
- Fix the script to support both Python 2 and 3 (pyFAI-calib, pyFAI-benchmark)
- Fix selection of units on diff-map (the user selection was not propagated)
- For users
- More source code in MIT license
- Update name and specification for cameras
- Add cameras: Eiger500k, RaspberryPi5M, RaspberryPi8M
- Fix Xpad S540 flat detector geometry
- Fix definition of CeO2 calibrant
- Add mask and flat on multi-geometry
- Fix solid angle of the multi-geometry
- Fix geometry processing for custom output space
- Fix normalization factor and variance
- Add support for Qt5
- Add support for Debian 9 packaging
- For developers
- Create common preprocessing for distortion correction
- Create common image preprocessing using Cython (NaN filter, flatfield, dark, polarisation)
- Refactoring of units module. It allows to register custom units.
- Worker can now use Writer
- Worker polarization argument is renamed into polarization_factor
- Remove the dependency from python-fftw3, use numpy instead
- Remove QtWebKit dependency
- Fix un-correction of images using sparse matrix from scipy
0.12.0: 06/06/2016¶
- Continuous integration on linux, windows using Python 2.7 and 3.4+
- Drop support of Python 2.6, 3.2, 3.3 and debian6 packaging
- New radial output units: Reciprocal spacing squared and log(q) ID02
- GPU accelerate version of ai.separate (Bragg & amorphous) ID13
- Quantile filtering in pyFAI-average ID02
- New graphical application for diffraction imaging ID21
- Migrate to a common structure with silx (reorganize tests, benchmarks, …)
- Extensions (binary sub-modules) have all been moved to ext directory
- Many improvements multigeometry integrators
- Compatibility with the copy module (copy.deepcopy) for azimuthal integrator ID02
- Distortion correction works also for non-contiguous detectors
- Update documentation and provide advanced tutorials:
- Introduction to pyFAI using the jupyter notebook
- detector calibration ID15, BM02
- Correction of detector distortion, examples of pixel detectors.
- calibrant calculation ID30
- error handling ID02, BM29
- pyFAI-integrate can now be used with or without GUI
- Many new detectors (ADSC, Pilatus CdTe, Apex II, Pixium):
- support for non-flat/curved detectors (Aarhus)
- non-contiguous detectors (WOS Xpad)
- Include tests and benchmarking tools as part of the library
- Better testing.
0.11.0: 07/2015¶
- All calibrant from NIST are now available, + Nickel, Aluminum, … with bibliographic references
- The Cell class helps defining new calibrants.
- OpenCL Bitonic sort (to be integrated into Bragg/Amorphous separation)
- Calib is available from the Python interface (procedural API), not only from the shell script.
- Many new options in calib for reset/assign/delete/validate/validate2/chiplot.
- reset: set the detector, orthogonal, centered and at 10cm
- assign: checks the assignment of groups of points to rings
- delete: remove a group of peaks
- validate: autocorrelation of images: error on the center
- validate2: autocorrelation of patterns at 180° apart: error on the center function of chi
- chiplot: assesses the quality of control points of one/multiple rings.
- Fix the regression of the initial guess in calib (Thanks Jon Wright)
- New peak picking algorithm named “watershed” and based on inverse watershed for ridge recognition
- start factorizing cython regridding engines (work ongoing)
- Add “–poni” option for pyFAI-calib (Thanks Vadim Dyakin)
- Improved “guess_binning”, especially for Perkin Elmer flat panel detectors.
- Support for non planar detectors like Curved Imaging plate developped at Aarhus
- Support for Multi-geometry experiments (tested)
- Speed improvement for detector initialization
- better isotropy in peak picking (add penalization term)
- enhanced documentation on http://pyfai.readthedocs.org
0.10.3: 03/2015¶
- Image segmentation based on inverse watershed (only for recalib, not for calib)
- Python3 compatibility
- include testimages into distribution
0.10.2: 11/2014¶
- Update documentation
- Packaging for debian 8
0.10.1: 10/2014¶
- Fix issue in peak-picking
- Improve doc & manpages
- Compatibility with PyMca5
0.10.0: 10/2014¶
- Correct Caglioti’s formula
- Update tests and OpenCL -> works with Beignet and pocl open source drivers
- Compatibility with MacOSX and windows
0.9.4: 06/2014¶
- include spec of Maxwell GPU
- fix issues with intel OpenCL icd v4.4
- introduce shape & max_shape in detectors
- work on marchingsquares/sorted controurplot for calibration
- Enforce the use the Qt4Agg for Matplotlib and other GUI stuff.
- Update shape of detector in case of binning
- unified distortion class: merge OpenCL & OpenMP implementation #108
- Benchmarks for distortion
- Raise the level to warning when inverting the mask
- set of new ImXpad detectors Related issue #111
- Fix issue with recalib within MX-calibrate
- saving detector description in Nexus files issue #110
- Update some calibrants: gold
- about to make peak-picking more user-friendly
- test for bragg separation
- work on PEP8 compliance
- Do not re-cythonize: makes debian package generation able to benefit from ccache
- conversion to SPD (rotation is missing)
- pixelwise worker
- correct both LUT & OCL for memory error
- replace os.linsep with “n” when file file opened in text mode (not binary)
- rework the Extension part to be explicit instead of “black magic” :)
- implement Kahan summation in Cython (default still use Doubles: faster)
- Preprocessing kernel containing all cast to float kernels #120
- update setup for no-openmp option related to issue #127
- Add read-out mode for mar345 as “guess_binning” method for detector. Also for MAR and Rayonix #125
- tool to benchmark HDF5 writing
- try to be compatible with both PySide and PyQt4 … the uic stuff is untested and probably buggy #130
- Deactivate the automatic saturation correction by default. now it is opt-in #131
0.9.3: 02/2014¶
- Better control for peak-picking (Contribution from Gero Flucke, Desy)
- Precise Rayonix detectors description thanks to Michael Blum
- Start integrating blob-detection algorithm for peak-picking: #70
- Switch fron OptParse to ArgPrse: #83
- Provide some calibrant by default: #91
- Description of Mar345 detector + mask#92
- Auto-registration of detectors: #97
- Recalib and check-calib can be called from calib: #99
- Fake diffraction image from calibrant: #101
- Implementation of the CSR matrix representation to replace LUT
- Tight pixel splitting: #43
- Update documentation
0.9.2: (01/2014)¶
- Fix memory leak in Cython part of the look-up table generation
- Benchmarks with memory profiling
0.9: 10/2013¶
- Add detector S140 from ImXpad, Titan from Agilent, Rayonix
- Fix issues: 61, 62, 68, 76, 81, 82, 85, 86, 87
- Enhancement in LImA plugins (better structure)
- IO module with Ascii/EDF/HDF5 writers
- Switch some GUI to pyQtGraph in addition to Qt
- Correction for solid-angle formula
0.8: 10/2012¶
- Detector object is member of the geometry
- Binning of the detector, propagation to the spline if needed
- Detector object know about their masks.
- Automatic mask for some detectors like Pilatus or XPad
- Implementation of sub-pixel position correction for Pilatus detectors
- LUT implementation in 1D & 2D (fully tested) both with OpenMP and with OpenCL
- Switch from C++/Cython OpenCL framework to PyOpenCL
- Port opencl code to both Windows 32/64 bits and MacOSX
- Add polarization corrections
- Use fast-CRC checksum on x86 using SSE4 (when available) to track array change on GPU buffers
- Support for flat 7*8 modules Xpad detectors.
- Benchmark with live graphics (still a memory issue with python2.6)
- Fat source distribution (python setup.py sdist –with-test-images) for debian
- Enhanced tests, especially for Saxs and OpenCL
- Recalibration tool for refining automatically parameters
- Enhancement of peak picking (much faster, recoded in pure Cython)
- Easy calibration for pixel detector (reconstruction of inter-module space)
- Error-bar generation using Poisson law
- Unified programming interface for all integration methods in 2theta, q or radius unit
- Graphical interface for azimuthal integration (pyFAI-integrate)
- Lots of test to prevent non regression
- Tool for merging images using various method (mean, median) and with outlayer rejection
- LImA plugin which can perform azimuthal integration live during the acquisition
- Distortion correction is available alone and as LImA plugin
- Recalibration can refine the wavelength in addition to 6 other parameters
- Calibration always done vs calibrant’s ring number, lots of new calibrant are available
- Selection by hand of single peaks for calibration
- New detectors: Dexela and Perkin-Elmer flat panel
- Automatic refinement of multiple images at various geometries (for MX)
- Many improvements requested by ID11 and ID13
0.7.2: 08/2012¶
- Add diff_tomo script
- Geometry calculation optimized in (parallel) cython
0.7: 07/2012¶
Implementation of look-up table based integration and OpenCL version of it
0.6: 07/2012¶
- OpenCL flavor works well on GPU in double precision with device selection
0.5: 06/2012¶
- Include OpenCL version of azimuthal integration (based on histograms)
0.4: 06/2012¶
- Global clean up of the code regarding options from command line and better design
- Correct the orientation of the azimuthal angle chi
- Rename scripts in pyFAI-calib, pyFAI-saxs and pyFAI-waxs
0.3: 11/2011¶
- Azimuthal integration splits pixels like fit2d
0.2: 07/2011¶
- Azimuthal integration using cython histogramming is working
0.1: 05/2011¶
- Geometry is OK