Page MenuHomePhabricator
Create Task
Maniphest T12287

[Dashboard] mitkImageVtkMapper2DLevelWindowTest fails on some machines
Closed, ResolvedPublic
Actions

Description

The test renders a sagittal slice of Pic3D and adapts the level window to use the "blood" preset. Afterwards the rendered result is compared to a reference image generated on the PC of Alexandra.

The test reproducible fails on:
MBI032-VM-NINST
MBI076

with the following output:
http://cdash.mitk.org/testDetails.php?test=3219338&build=145863
Note: the errors are smaller then the pixel-size of the input image!!!

The test also fails on:
MBI077
MBI029

with no output (I assume that this will occur randomly):
http://cdash.mitk.org/testDetails.php?test=3220824&build=145977

The test passes on:
mes-macmini02.local
mbimac05.inet.dkfz-heidelberg.de
MBI013
MBI086
MBI084
MBI025
MBI071

We should try to figure out what differences are between the passing and the failing dartclients.

Theories for the reasons are:
-0.5er offset during reslicing
-OpenGL uncertainties on different machines

Event Timeline

• kilgus added projects: MITK, Bugzilla.Jun 22 2012, 9:46 AM
• goch added a subscriber: • goch.Jun 22 2012, 10:46 AM

MBI032-VM-NINST is a virtual machine with the following specifications:

Microsoft Windows XP Professional Version 2002 Service Pack 3
Intel(R) Core(TM) i7-3930K CPU @ 3.20 GHz 3.18 GHr, 3,00 GB RAM

Graphic card:
VirtualBox Graphics Adapter

• goch added a comment.Jun 22 2012, 10:58 AM

The vm host server uses a GeForce GTX 560

• schroedt added a subscriber: • schroedt.Jun 22 2012, 10:58 AM

The results from the dartclients seem to be equal but a part of the image (Alfred) is somehow shifted. This is what gimp shows as diff. (level window adjusted)

DiffAlfredCaspar.png (480×640 px, 2 KB)

• kilgus added a subscriber: • kilgus.Jun 22 2012, 11:03 AM

We figured out that the two machines
MBI032-VM-NINST
MBI076
render the pixels with the sequence of the size 6,6,6,5 resulting in a total of 64 pixels which is incorrect.

The reference image uses the sequence 8,7,8,7 resulting in a total of 49 pixels, which is the correct amount for Pic3D sagittal.

New theory list:
Theories for the reasons are:
-0.5er offset during reslicing -> rather not the case. would happen on more machines
-OpenGL uncertainties on different machines
-Hardware/Software modus interpolation!

• kilgus added a comment.Jun 22 2012, 11:04 AM

Todo: We still have no idea why this only happens in combination with the level window test.

franza added a subscriber: franza.Jun 22 2012, 12:44 PM

configuration of MBI076:

Windows 7
64 bit
Quad-Core
ATI Radeon HD 5700

• kilgus added a comment.Jul 26 2012, 2:09 PM

Todo:
add a software rendered reference image to the data repository and reactivate the test

• kilgus added a comment.Aug 7 2012, 3:03 PM

The test is working now on most machines. The problems mentioned here do not occur with VTK 5.8.

kislinsk edited projects, added MITK (2012-09); removed MITK.Aug 1 2016, 10:48 PM
MITK · MIC · IMSY · DKFZ