[Sticky] Astro Pixel Processor 2.0.0-beta45 release notes

Astro Pixel Processor 2.0.0-beta45 release notes

This release has many fixes, improvements and new optimizations especially for the new registration engine that was introduced in 2.0.0-beta39. And fully multi-threaded Local Normalization Correction.

• UPGRADED development platform to OpenJDK 26.0.1

  We moved from Oracle to OpenJDK because of the next release note, to further improve the internal memory configuration and cleanup. 2.0.0-beta45 is using the Shenandoah adaptive garbage collector (developed mainly by the open source community and Red Hat) which is again an improvement over the new configuration in beta43 and beta44 that were using the Z garbage collector.

• IMPROVED memory configuration

  We have further tuned the internal memory configuration to prevent Out Of Memory and Memory fragmentation on big datasets as reported here: https://www.astropixelprocessor.com/community/main-forum/encountered-error-in-module-createqualityregistrationdescriptorstask/#post-34515 We are now using adaptive Shenandoah garbage collection in the java runtime. Memory should be released back to the Operating System more quickly than with beta43 and beta44 while maintaining processing speed.

• IMPROVED ALGEBRA, MATRIX INVERSION - performance improvement !

  We used to rely on a third party algebra library  (EJML) to do matrix inversion. The library could do it quite fast but only single threaded. This was the last remaining issue for getting fully multi-threaded Local Normalization Correction calculations for example. We have now implemented our own very fast single thread matrix inverter which out runs the EJML matrix inversion on small matrices with a factor of almost 10. And ! we have also implemented a multi-threaded matrix inversion which really helps to improve performance in both the registration engine and in speeding up Local Normalization Correction on big datasets.

• IMPROVED LOCAL NORMALIZATION CORRECTION, now 100% multi-threaded.

  As a result of our new multi-threaded matrix inversion that we implemented. LNC is now fully multi-threaded and so this is a nice performance gain for anyone using LNC.

• IMPROVED 4)REGISTER, 2-view pattern recognition

  We have optimized all the code in the 2-view registration phase in terms of memory usage. The total memory consumption is grealy reduced while registering images, which should enhance performance when registering 100s of frames. We have also fixed a couple of minor inconsistencies in the engine that could have caused variable registration results, so variance is reduced meaning more robust results.

• IMPROVED 4)REGISTER, mosaic descriptor creation

  When running a very big mosaic (100+ frames) APP used to create all the descriptors for each frame first and then start registering the frames to each other. This implementation was rather memory hungry because all the descriptors were needed to stay in memory. It could cause systems to run Out Of Memory https://www.astropixelprocessor.com/community/main-forum/encountered-error-in-module-createqualityregistrationdescriptorstask/#post-34515 . Beta45 will now only create the descriptors for a reference frame when needed and will remove it from memory immediately when it is no longer needed. In this way, memory consumption stays under control and the performance is not affected and might be slightly better on certain systems with big mosaics.

• IMPROVED 4)REGISTER, pattern recognition will be faster and use less memory

  We have made several optimizations in the 2-view registration phase where patterns between the stars are matched between the frames to be registered. The K-dimensional tree implementation was upgraded to be faster and use less memory.

• FIXED 4)REGISTER dynamic distorion correction was disabled on pierflipped data

  As mentioned here https://www.astropixelprocessor.com/community/main-forum/app-2-0-beta-41-registration-error/#post-34353 registration did not work properly on frames that were pierflipped. beta39-beta44 (new registration engine) mistakenly turned off dynamic distortion correction on these frames, but they clearly need it. This has been robustly fixed by adding an algorithm in the registration engine that detects a pier flip. 

• FIXED 4)REGISTER no more force of calibrated projective model for data that is near 200mm in focal length.

  As reported here https://www.astropixelprocessor.com/community/appreleases/beta-41-registration-issues/#post-34210 if the projective registration model was used on data that has a focal length near 200mm, the engine could warn the user that the calibrated projective model would be needed and it would force that model. That behaviour has been changed. The user will be asked now to continue with the projective model or if he/she wants to change the model. VIce versa also for starting with the calibrated projective model and possibly using the projective model when the mosaic calculation sees that projective is also viable.

• FIXED 4)REGISTER comet/one-star 

  As reported here https://www.astropixelprocessor.com/community/comet-registration-and-integration/beta-41-comet-registration-issues/#post-34307 the comet/one-star regisgtration mode had several issues with data that has little stars and low quality (shot in the morning glow). We have fixed 4 problems so now the data can be properly processed. We have fixed 2 bugs related to little stars so the images might not be able to register on the stars. We fixed the counter of which image is being processed in the comet mode and we also added a warning message if star registration is simply not possible and the user should try to disable star analysis and the register on the comet only.

• FIXED SAVE/LOAD Preferences

  Several issues have been fixed for saving and loading the user preferences between APP restarts. For instance, if APP starts with a crash for some reason, or if APP start is aborted before being fully loaded, the next APP instance could start with not default settings, like in 3) Analyse Stars, it could start with max stars of only 10. This has been fixed. If such an APP crash/abort occurs, APP will detect it the next time it is started and it will safely load all the default settings. 

• FIXED ArrayIndexOutOfBoundsException in LokationAndScaleCalculator for NormalizationMap