这是indexloc提供的服务,不要输入任何密码
Skip to content

Conversation

@sboukortt
Copy link
Member

Previously, the Butteraugli loop would use an intensity of 80 even for HDR content, thereby lowering quality at efforts 8 and above.

butteraugli_main also did not handle the reference and actual images having different intensity_targets, meaning that it was not possible to compare a PQ PNG (which would automatically have an intensity_target of 10000) with a compressed JXL version using a lower intensity_target (reflecting the fact that the image doesn’t actually get that bright).

ClassE_LasVegasStore, before:

e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
e8:  1.415 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
e9:  1.413 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
e10: 1.409 bpp, Butteraugli 1.9175151587, 3-norm 0.770864

After:

e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
e8:  1.721 bpp, Butteraugli 1.4964444637, 3-norm 0.590536
e9:  1.727 bpp, Butteraugli 1.2693173885, 3-norm 0.585146
e10: 1.719 bpp, Butteraugli 1.3858509064, 3-norm 0.589227

Previously, the Butteraugli loop would use an intensity of 80 even for
HDR content, thereby lowering quality at efforts 8 and above.

butteraugli_main also did not handle the reference and actual images
having different intensity_targets, meaning that it was not possible to
compare a PQ PNG (which would automatically have an intensity_target of
10000) with a compressed JXL version using a lower intensity_target
(reflecting the fact that the image doesn’t actually get that bright).

ClassE_LasVegasStore, before:

    e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
    e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
    e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
    e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
    e8:  1.415 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
    e9:  1.413 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
    e10: 1.409 bpp, Butteraugli 1.9175151587, 3-norm 0.770864

After:

    e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
    e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
    e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
    e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
    e8:  1.721 bpp, Butteraugli 1.4964444637, 3-norm 0.590536
    e9:  1.727 bpp, Butteraugli 1.2693173885, 3-norm 0.585146
    e10: 1.719 bpp, Butteraugli 1.3858509064, 3-norm 0.589227
@sboukortt sboukortt added this pull request to the merge queue Oct 11, 2024
Merged via the queue into libjxl:main with commit 095f259 Oct 11, 2024
101 of 102 checks passed
@sboukortt sboukortt deleted the butteraugli-intensity-target branch October 11, 2024 16:11
mo271 pushed a commit to mo271/libjxl that referenced this pull request Nov 26, 2024
Previously, the Butteraugli loop would use an intensity of 80 even for
HDR content, thereby lowering quality at efforts 8 and above.

butteraugli_main also did not handle the reference and actual images
having different intensity_targets, meaning that it was not possible to
compare a PQ PNG (which would automatically have an intensity_target of
10000) with a compressed JXL version using a lower intensity_target
(reflecting the fact that the image doesn’t actually get that bright).

ClassE_LasVegasStore, before:

    e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
    e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
    e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
    e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
    e8:  1.415 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
    e9:  1.413 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
    e10: 1.409 bpp, Butteraugli 1.9175151587, 3-norm 0.770864

After:

    e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
    e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
    e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
    e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
    e8:  1.721 bpp, Butteraugli 1.4964444637, 3-norm 0.590536
    e9:  1.727 bpp, Butteraugli 1.2693173885, 3-norm 0.585146
    e10: 1.719 bpp, Butteraugli 1.3858509064, 3-norm 0.589227

(cherry picked from commit 095f259)
mo271 pushed a commit that referenced this pull request Nov 26, 2024
Previously, the Butteraugli loop would use an intensity of 80 even for
HDR content, thereby lowering quality at efforts 8 and above.

butteraugli_main also did not handle the reference and actual images
having different intensity_targets, meaning that it was not possible to
compare a PQ PNG (which would automatically have an intensity_target of
10000) with a compressed JXL version using a lower intensity_target
(reflecting the fact that the image doesn’t actually get that bright).

ClassE_LasVegasStore, before:

    e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
    e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
    e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
    e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
    e8:  1.415 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
    e9:  1.413 bpp, Butteraugli 1.9081192017, 3-norm 0.770197
    e10: 1.409 bpp, Butteraugli 1.9175151587, 3-norm 0.770864

After:

    e1:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e2:  1.709 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e3:  1.616 bpp, Butteraugli 1.9095990658, 3-norm 0.763096
    e4:  1.631 bpp, Butteraugli 1.9226754904, 3-norm 0.761637
    e5:  1.519 bpp, Butteraugli 2.5887553692, 3-norm 0.801795
    e6:  1.507 bpp, Butteraugli 2.4450392723, 3-norm 0.777965
    e7:  1.681 bpp, Butteraugli 1.4998193979, 3-norm 0.617420
    e8:  1.721 bpp, Butteraugli 1.4964444637, 3-norm 0.590536
    e9:  1.727 bpp, Butteraugli 1.2693173885, 3-norm 0.585146
    e10: 1.719 bpp, Butteraugli 1.3858509064, 3-norm 0.589227

(cherry picked from commit 095f259)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants