Ticket #43 (closed defect: invalid)

Opened 11 years ago

Last modified 11 years ago

xuvtools stitches with bad score when not using full search

Reported by: emmenlau Owned by: emmenlau
Priority: major Milestone:
Component: xuvtools Version:
Keywords: stitcher, xuvtools, error Cc: aaron@…

Description

Report by AP: From my/our testing stitching from known coordinates almost never works, even when starting from very close from the correct solution. I compared the log files (attached) of two runs, one with comprehensive search and one starting using the coordinates from the solution of the comprehensive stitch with just minor manual shifts of the tiles. It looks to me that in the "coordinates"-case the overlaps are found okay, and the correct tiles are tested against each other. The obvious difference between the two stitching modalities is the "pairwise score".

Attachments

logs.tar Download (6.7 KB) - added by aaron@… 11 years ago.
Augmented logs (with additional information to order log data to function calls, good for diffs) to compare the behavior of full stitch from stitch with coords. Data set was generated with the MATLAB script = XuvGenHDF5DisplacedCircles('XuvTest?-DisplacedCircles?_0-32-32.h5', [32 32], [128 128 8]). The tiles were placed on top of each other for the COORDS mode (full overlap).

Change History

Changed 11 years ago by emmenlau

  • cc aaron@… added
  • reporter changed from mario@… to emmenlau

Full search:

(line 1148) pairwiseScore is (starting from 1, not 0):

0.000 0.634 0.678 0.090 0.000 0.000 0.000 0.000 0.716 0.000 0.000 0.000 0.000 0.000 0.000 0.700 0.000 0.000 0.000 0.000 0.000 0.000 0.748 0.115 0.000 0.000 0.000 0.000 0.000 0.631 0.000 0.000 0.000 0.000 0.000 0.000

and this then becomes:

(line 1209) connected components-test found for thresh 0.5: (0,5) x (0,5) [ 0 0.634428 0.677591 0.089841 0 0

0 0 0.716484 0 0 0 0 0 0 0.700313 0 0 0 0 0 0 0.748211 0.114698 0 0 0 0 0 0.631252 0 0 0 0 0 0 ]

0 1 2 3 4 5 1 1 1 1 1 1

Using coordinates:

(line 742) pairwiseScore is (starting from 1, not 0):

0.000 0.059 0.599 0.090 0.000 0.000 0.000 0.000 0.085 0.000 0.000 0.000 0.000 0.000 0.000 0.095 0.000 0.000 0.000 0.000 0.000 0.000 0.743 0.115 0.000 0.000 0.000 0.000 0.000 0.461 0.000 0.000 0.000 0.000 0.000 0.000

and this then becomes:

(line 767) connected components-test found for thresh 0.5: (0,5) x (0,5) [ 0 0.0585024 0.599137 0.089841 0 0

0 0 0.085424 0 0 0 0 0 0 0.0945465 0 0 0 0 0 0 0.743306 0.114698 0 0 0 0 0 0.461172 0 0 0 0 0 0 ]

0 1 2 3 4 5 2 1 2 4 4 5

Changed 11 years ago by emmenlau

Test dataset "To151007 40xma":

Bug0043/To151007 40xma001.ims
Bug0043/To151007 40xmb002.ims
Bug0043/To151007 40xmc003.ims
Bug0043/To151007 40xmd004.ims
Bug0043/To151007 40xme005.ims
Bug0043/To151007 40xmf006.ims

gives the following (different) results for using / not using phaseonly correlation:

fineDisplacement for tiles   0,   1, phase-test  0 is (0,-1,0)   	with score 0.634428	with phaseonly corr 
fineDisplacement for tiles   0,   1, phase-test  0 is (-1,1,-2)   	with score 0.61722	without phaseonly corr 
fineDisplacement for tiles   0,   2, phase-test  0 is (-1,0,0)   	with score 0.677591	with phaseonly corr 
fineDisplacement for tiles   0,   2, phase-test  0 is (3,1,-4)   	with score 0.660634	without phaseonly corr 
fineDisplacement for tiles   0,   3, phase-test  0 is (1,-16,-16)   	with score 0.089841	with phaseonly corr 
fineDisplacement for tiles   1,   2, phase-test  0 is (-1,-1,1)   	with score 0.716484	with phaseonly corr 
fineDisplacement for tiles   1,   2, phase-test  0 is (3,0,-3)   	with score 0.461036	without phaseonly corr 
fineDisplacement for tiles   2,   3, phase-test  0 is (1,0,1)   	with score 0.700313	with phaseonly corr 
fineDisplacement for tiles   2,   3, phase-test  0 is (-4,-2,-5)   	with score 0.166809	without phaseonly corr 
fineDisplacement for tiles   3,   4, phase-test  0 is (0,0,-1)   	with score 0.748211	with phaseonly corr 
fineDisplacement for tiles   3,   4, phase-test  0 is (-4,2,-1)   	with score 0.59466	without phaseonly corr 
fineDisplacement for tiles   3,   5, phase-test  0 is (3,15,6)   	with score 0.114698	with phaseonly corr 
fineDisplacement for tiles   4,   5, phase-test  0 is (1,0,0)   	with score 0.631252	with phaseonly corr 
fineDisplacement for tiles   4,   5, phase-test  0 is (-2,-1,1)   	with score 0.624473	without phaseonly corr 

Changed 11 years ago by aaron@…

Augmented logs (with additional information to order log data to function calls, good for diffs) to compare the behavior of full stitch from stitch with coords. Data set was generated with the MATLAB script = XuvGenHDF5DisplacedCircles('XuvTest?-DisplacedCircles?_0-32-32.h5', [32 32], [128 128 8]). The tiles were placed on top of each other for the COORDS mode (full overlap).

Changed 11 years ago by emmenlau

  • status changed from new to closed
  • resolution set to invalid

Error was in stage coordinates.

Note: See TracTickets for help on using tickets.