#99 2.1.0 Regression, Black corner with Mantiuk06 contrast eq

Luminance_HDR_2.1.0
open
Davide
Program (114)
5
2013-05-17
2011-08-30
Naggobot
No

Sorry if this comes as a duplicate. I am redoing this since it seems that the first was unsuccessful.

System: Ubuntu 10.04 amd 64.
When tonemapping using Mantiuk06 contrast equalization a black right hand lower corner is produced.
This is a regression 2.0.1 and 2.0.2 do not suffer from this and when tonemapping a HDR created with 2.1.0 with 2.0.1 or 2.0.2 a black corner is not produced. Also 2.1.0 produces otherwise lighter image except the black corner on right hand bottom.

I downgraded to both 2.0.1 and 2.0.2 to test this. 2.0.1 was from http://ppa.launchpad.net/ferramroberto/linuxfreedomlucid/ubuntu
and 2.0.2 from https://launchpad.net/~dhor/+archive/myway/+build/2490209
2.1.0 was from
http://ppa.launchpad.net/dhor/myway/ubuntu

I try to get the tonemapped images available through ubuntuone but it seems to take a long time to upload / publish. I continue tomorrow. I do not want to add the attachment here in case it was the reason the previous one did not go through.

Discussion

1 2 > >> (Page 1 of 2)
  • Naggobot
    Naggobot
    2011-10-29

    I just tested latest build from https://launchpad.net/~dhor and unfortunately the black corner is still there.

     
  • Toni
    Toni
    2012-05-28

    Happens also with version 2.2.1. Although the black area is smaller than what 2.0.0 had.
    Version 2.0.2 shows no black area also.

    Also one observation is that when this happens in 2.2.1, the console output gets this warning about iterations:

    pfstmo_mantiuk06 (Mode: Contrast Equalization, scaleFactor: -0.1, saturationFactor: 0.8, detailFactor: 1)
    pfstmo_mantiuk06: Warning: Not converged (hit maximum iterations), error = 0.141769 (should be below 0.001).

    And the progress bar in GUI doesn't ever reach to 100%, but goes back and worth near between 0-41%.
    I guess the algorithm never quite reaches the lower right corner, which causes it to be left black.

    Maybe this is related:
    I also downloaded the pfstools (1.8.3) and pfstmo (1.4), and there is exactly same issue with those tools:

    $pfsin test.exr | pfstmo_mantiuk06 -v -e 0.3 | pfsout out.tiff
    pfstmo_mantiuk06: algorithm: contrast equalization
    pfstmo_mantiuk06: contrast scale factor = 0.3
    pfstmo_mantiuk06: saturation factor = 0.8
    pfstmo_mantiuk06: using conjugate gradients (itmax = 200, tol = 0.001).
    completed 22%%
    pfstmo_mantiuk06: Warning: Not converged (hit maximum iterations), error = 0.166611 (should be below 0.001).

    ... and the out.tiff has black lower right corner.

    Update 25.12.2012:
    Mac OSX 10.8.2 and luminance 2.3.0, same black lower right corner...

     
    Last edit: Toni 2012-12-25
  • Naggobot
    Naggobot
    2012-12-29

    Sorry to have been so quiet about this for so long. I have been sticking with 2.0.2 due to this issue but since Lucid support is ending soon I will have to update the OS and possibly Luminance also. Between versions 2.0.2 and 2.1.0 files /src/TonemappingOperators/mantiuk06/contrast_domain.cpp and pfstmo_mantiuk06.cpp have changed in size i.e. it is reasonable to assume that these have been changed and that these changes may have caused the regression between 2.0.2 and 2.1.0? (note the question mark). In any case the diff outputs are attached. There are not much changes but these are total gibberish to be.

     
    Attachments
  • Davide
    Davide
    2012-12-29

    There is little difference between these implementations of the Mantiuk06 operator. However, the upcoming 2.3.1 will have a quite different implementation (deeply inspired by the old one though). It would be nice to have the HDR image that you are using to generate the regression

     
  • Davide
    Davide
    2012-12-29

    • assigned_to: Giuseppe Rota --> Davide
    • milestone: --> Luminance_HDR_2.1.0
     
  • Toni
    Toni
    2012-12-29

    You can find HDR file, screenshot of luminance showing the settings, and sample output file from here: http://bleach.kapsi.fi/sample.tar.gz

    OS: Mac 10.8.2
    Luminance: 3.2.0 build be5409

     
  • Naggobot
    Naggobot
    2013-02-09

    Had to finally start upgrading my ubuntu installation. On Ubuntu 12.04 Precise Pangolin Luminance HDR 2.3.0 still has this regression. If started from terminal following error is output:

    pfstmo_mantiuk06: Warning: Not converged (hit maximum iterations), error = 0.240958 (should be below 0.001).
    Monitor profile: ""
    Printer profile: ""

     
  • at least on windows even the 2.0.2 version produces exactly the same result as the 2.3.0 does, including the black corners

     
1 2 > >> (Page 1 of 2)