Automatic Perceptual Dedupe reports don't get created

[Ricardo50]

Describe the bug Automatic Perceptual Dedupe reports don't get created when uploading an image similar to the ones that is already there.

To Reproduce Steps to reproduce the behavior:

1. Upload an image to the board
2. Upload a second image when the first one is done processing. It can be of a different resolution, with a small difference or with a different file format

Expected behavior I expect to see an automatic dedupe report, but none get made

Screenshots If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: Windows 10 64-bit
• Browser: Firefox 64-bit
• Version: 81

Additional context I've tested with many different image pairs and options

[basisbit]

Thank you for creating this bug report and thus helping improve the imageboard software! Could you please attach the two sample images to the issue, so that we can try to reproduce the problem? Also, how did you set up the imageboard - is it running as docker containers, is it running in production mode or dev mode (see your .env file) and what version of the imageboard are you running (which git commit or clone date)?

[Ricardo50]

Thank you for the quick response. I haven't been sitting still and been testing for a while. See the attached sample images below:

What I found out is that if I set the option duplication_checked for the other image of the duplicate pair to false and then perform an image repair on the other similar image that an duplicate report will be made

If I change the code lib/philomena/duplicate_reports.ex, line 20 to |> where([i, _it], i.duplication_checked == true) (aka inverting the check), it seems to work.

What I found with this basic troubleshoot is that the logic is probably inverted. Is this helpful?

[basisbit]

The feature itself seems to be working fine with the booru fork. The job runs and perceptual deduplication reports are being created. For example the images https://manebooru.art/images/614170 (english) and https://manebooru.art/images/614171 (spanish) are reported for being possibly duplicates. Of course, neither your sample images nor my sample images are duplicates, but rather have "easily" visible changes that even can be seen on a 128x128 pixel scaled down image.

Regarding the code change that you suggest, I'd guess that @liamwhite would be the best to ask.

I can confirm that the sample images which you provided do not cause philomena to create a perceptive duplication report.