This post belongs to a parent (learn more) « hide
Resized to 70% of original (view original)
takatsuki yayoi (idolmaster) drawn by tokiani

Artist's commentary

  • Original
    •

    アイマス　Twitterまとめ

    Twitterに上げた物をまとめました。
    冬コミ3日目にスペース頂きましたのでそのうち告知上げさせて頂きます。

    ミキミキ誕生日おめでとう！

  • Comments
  • Recommended
    •
    nonamethanks
    [hidden]

    You'll have to ask @ZipFile that, since he made that site.

    Verifying with numpy I get that the two pictures are identical even though the signature is different.

    >>> import cv2, numpy as np
>>> im1 = cv2.imread("7b15f7529ab1af57464549048bf8d382.png")
>>> im2 = cv2.imread("8a81ddb788a46f8b036ce59659b7b150.png")
>>> np.sum(im1 != im2)
0
    $ optipng -verbose -o1 -zs0 -f0 8a81ddb788a46f8b036ce59659b7b150.png -out 1.png
$ optipng -verbose -o1 -zs0 -f0 7b15f7529ab1af57464549048bf8d382.png -out 2.png
>>> import cv2, numpy as np
>>> im1 = cv2.imread("1.png")
>>> im2 = cv2.imread("2.png")
>>> np.sum(im1 != im2)
0
  • 0
  • Reply
    • ZipFile
    [hidden]

    Verified manually - both images are pixel-identical.

    Duplicatebooru uses optipng to strip redundant alpha channels (they mess with image hash calculation), but post #2161300 hit one specific edge-case when optipng won't do anything. So resulting hashes were different for both images, hence not reported as identical.

    nerd stuff

    Duplicatebooru (for each image) runs optipng with -o1 -zs0 -f0 params to speed up things and produce stable result. Output of optipng is then passed to imagemagick to calculate image hashes. If two hashes match - images reported as duplicates.

    Exact problem is reproducible like this: optipng 8a81ddb788a46f8b036ce59659b7b150.png 7b15f7529ab1af57464549048bf8d382.png -o1 -zs0 -f0. What's going on here is that optipng tries to strip redundant alpha channels from both images, but the 7b15f7529ab1af57464549048bf8d382.png produces exact the same image size, which means optimization was useless (from optipng point of view) and does not modify the image. So the result is one image with stripped alpha and one untouched, which results in different image hashes.

  • 0
  • Reply