Author Topic: TVGDenoise Corrupts Image  (Read 1222 times)

Offline GordonH

  • Newcomer
  • Posts: 27
TVGDenoise Corrupts Image
« on: 2019 January 31 12:08:38 »
TVGDenoise is corrupting my image. It worked fine in a preview, but, when I applied it to the image a part of the image turned black. I tried rebooting my PC and reloading from a project with the same result. I tried it in RGB/K mode and in CIE L*a*b* mode with the same results. See attachment

The image is a 64 bit if that matters. PixInsight is up todate Version 01.08.06.1457

Thanks,
Gordon Hansen

Offline Juan Conejero

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 7111
    • http://pixinsight.com/
Re: TVGDenoise Corrupts Image
« Reply #1 on: 2019 February 01 01:26:20 »
Hi Gordon,

Wow, this one is spectacular for sure. I cannot reproduce it with the images I have at hand. Can you please upload this particular image, just as you are using it to cause this problem? I'll need also the process instance. If you can write a bundled project (.pxiproject folder) with the image and the TGVDenoise tool open and loaded with your parameters, that would be great. Thank you!
Juan Conejero
PixInsight Development Team
http://pixinsight.com/

Offline GordonH

  • Newcomer
  • Posts: 27
Re: TVGDenoise Corrupts Image
« Reply #2 on: 2019 February 01 07:47:25 »
I created the project. Its 29.6GB. The "Attach" option will not accept the project - I think because its a directory not a file. Is there another way of uploading the project?

I re-ran TBGDenoise again this morning. Windows did an update overnight and I wanted to make sure this wasn't an operating system bug. I re-ran on the preview and it converged after 142, 186, & 171 iterations. When run on the full image it never converged and went the full 1500 iterations set for. I'm not aware of that ever happening before. I had run TVGDenoise earlier in my processing with no problem, but, I did some more stretching and wasn't happy with the noise in the now brighter areas. The convergence value was still set to the default .004 level.

I will re-run at a higher limit to see what happens.

Thank you,
Gordon

Offline GordonH

  • Newcomer
  • Posts: 27
Re: TVGDenoise Corrupts Image
« Reply #3 on: 2019 February 01 14:50:47 »
I was mistaken earlier - the project size 6.08 GB.

I did go back and do some experimenting. I increased the convergence value from .004 to 1.0 with the same result. I then applied a inverted luminance mask on the image, again with the same result.

I then rolled back the processes to before the first TVGDenoise and then reapplied all the steps. Then applied the TVGDenoise (@.004 level) and all worked as expected. It appears that multiple applications of TVGDenoise is the problem.

If you still want me to upload the project I will. Let me know the best way.

Gordon

Offline pfile

  • PTeam Member
  • PixInsight Jedi Grand Master
  • ********
  • Posts: 4729
Re: TVGDenoise Corrupts Image
« Reply #4 on: 2019 February 01 16:10:04 »
you can just use google drive or dropbox or whatever and post a link here

rob

Offline GordonH

  • Newcomer
  • Posts: 27
Re: TVGDenoise Corrupts Image
« Reply #5 on: 2019 February 02 07:40:32 »

Offline KBALLZZ

  • Newcomer
  • Posts: 5
Re: TVGDenoise Corrupts Image
« Reply #6 on: 2019 February 02 14:58:24 »
Something similar has happened to me before (using MMT) and the problem was that my computer was maxed out in RAM usage. I saved the working image and started a new project (thus using less RAM) and it performed quickly without error. This may not be your specific issue, but thought I'd share.

Offline GordonH

  • Newcomer
  • Posts: 27
Re: TVGDenoise Corrupts Image
« Reply #7 on: 2019 February 04 11:24:53 »
Thanks for the info on RAM useage. I took a look by opening the Windows Resource Monitor to track RAM useage while TVGDenoise was not running compared to running. (See the graph attached.) On my system TVGDenoise barely shows up on the graph. You can just see the uptick in usage about 4 segments from the left in the graph.  Increase in RAM was about 1 MB. (My system has 24 MB so I'm have some headroom.

Gordon