About Dewarp - speed test and other different thoughts

[jourmager]

1. Speed ​​test at dewarp stage:
   PerspectiveTestSet
   Scan Tailor Deviant = 1 thread (15% CPU load) = 1 min 43 sec
   Scan Tailor experimental 2016-02-22 by Tulon = 8 threads (100% CPU load) = 21 sec
   WarpTestSet
   Scan Tailor Deviant = 1 thread (15%-30% CPU load) = 3 min 16 sec
   Scan Tailor experimental 2016-02-22 by Tulon = 8 threads (100% CPU load) = 34 sec

The speeds of Scan Tailor experimental 2016-02-22 by Tulon and scantailor-experimental-1.2024.11.18-Win32-X86-64-Qt6 are the same.

STD uses the STU architecture, which uses the 1-threaded architecture of the original ST, which was created in 2008.
ST experimental by Tulon (and STEX) uses multi-threaded architecture of 2016.

STA, STU, STexp were created almost simultaneously, so improvements in the STexp architecture (such as fix geometric distortions) could not be used in either STA or STU.

STD loses in speed by 5-7 times due to the use of 1-threaded architecture for processing geometric distortions.

Summary: until STD uses multi-threaded processing for processing geometric distortions, using STD to correct geometric distortions does not make practical sense.

2. about the test sets PerspectiveTestSet, WarpTestSet, etc
   These test sets use images with a gray checkerboard background with sharp transitions, as well as black fields around the test text.
   This is completely different from the images the algorithm was tested on and contradicts real-life cases.
   This leads to incorrect operation of the algorithms.
   After removing the checkerboard background and removing the black fields, the number of defective images decreased by 2 times.

3. Some of the images in the PerspectiveTestSet have distortions in several directions at once, while the option is called Keystoning - correction of trapezoidal distortions.
   The remaining defective images are just like that.

4. what ever...

Summary: don't look for a black cat in a dark room and don't complicate your own life.
Summary 2: if you want to make a program that people will use and who will be grateful to you for the program - then you can fix those elementary problems that have not been solved in 15 years in other ScanTailors.