Regarding WebP Lossless

[MasterInQuestion]

[[
[ honorificabilitudinitatibus@tutanota.com @ CE 2024-10-28 05:30:00 UTC:
    I noticed recently significant changes on the lossless things:
    https://github.com/webmproject/libwebp/commits?until=2024-10-09

    I researched several things in:
    https://github.com/webmproject/libwebp/commit/33357131698ba0abb5b7a32b6f274f6f58170642#Conclusion
    https://github.com/MasterInQuestion/talk/discussions/categories/WebP

    Would like your notice.
    Thanks. ]

----

[ jzern@google.com @ CE 2024-10-28 17:04:11 UTC:
    Yes, there have been changes recently to improve lossless compression density on a range of images.

    https://github.com/webmproject/libwebp/commit/33357131698ba0abb5b7a32b6f274f6f58170642#Conclusion
    https://github.com/MasterInQuestion/talk/discussions/categories/WebP
    .
    These don't look to be related to the RFC.
    If you have questions about lossless or near lossless please feel free to email: [email protected]
    Any issues can be filed at "issues.webmproject.org". ]

----

[ honorificabilitudinitatibus@tutanota.com @ CE 2024-10-29 04:49:27 UTC:
    Thanks for your attention.

    I don't yet have a Google account, and it might take me some effort to setup one.
    (tried in the past; but failed for tricky phone 2FA)
    Otherwise I should have forwarded relevant posts to:
    https://groups.google.com/a/webmproject.org/g/webp-discuss/c/0GmxDmlexek
    https://issues.webmproject.org/issues?q=componentid:1618983

    I also tried "[email protected]" months ago:
    It however seems to reject my email..?

    I mainly use GitHub. And in many cases more favored than email.
    Feel free to directly comment on my posts. (creating new queries as well)


    Side note:
    "issues.webmproject.org" looks awful on narrow screen. (1080x1920)
    (severe clipping of issue titles to almost unusable extent) ]

----

[ jzern@google.com @ CE 2024-10-31 02:26:48 UTC:
    Near lossless was never fully developed and likely could use some attention.
    That's part of the reason I suggested the mailing list or a bug to give it more visibility.
    This feature isn't widely used, and if it is: it might be in the 60 ~ 80 range.
    Patches to improve the behavior and better bridge the gap from lossy "-q 100" to "-lossless" would be accepted: though no one is actively working on this feature.


    There were some issues with some groups for a while earlier this year.
    Currently posting to the mailing list should work, though messages may be moderated and take a little while to appear.

    Thanks for the feedback.
    This was a recent migration, Chromium is using the same tracker.
    I'll look around to see if I can file a feature request. ]

----

[ honorificabilitudinitatibus@tutanota.com @ CE 2024-10-31 04:51:12 UTC:
    Significant research vouches for "-near_lossless 40":
    Not only my own previously mentioned. [1]
    Also sporadic reports by other people.
[ [1]
    Note GitHub's recent Broken Beta may cause glitchy anchor recognition of commit page.
    More details: https://github.com/orgs/community/discussions/139005#discussioncomment-10713581 ]

    I believe lossy ≈ pre-filtered lossless.
    Currently testing several concepts in the denoise techs.
    That would probably be eventually an independent general data filter.


    I'll try again later.
    But for [ https://groups.google.com/a/webmproject.org/g/webp-discuss/c/0GmxDmlexek ]:
    I think it's best to cross-link [ https://github.com/webmproject/libwebp/commit/33357131698ba0abb5b7a32b6f274f6f58170642#commitcomment-139962312 ] there?

    Currently I have difficulty access the issue tracker.
    But my relevant findings should reflect in:
    https://github.com/MasterInQuestion/talk/discussions/categories/WebP


    Previously I erroneously dropped the "[email protected]" CC in email recipient.
    Should I add it back for future messages? ]

----

[ jzern@google.com @ CE 2024-10-31 20:07:59 UTC:
    The thread is very old.
    If there's any discussion to have about near-lossless a new one would be better.

    No, I don't want to clutter that list.
    That is for mail related to the WebP format RFC, near-lossless is preprocessing and not a part of the bitstream. ]

----

[ honorificabilitudinitatibus@tutanota.com @ CE 2024-11-01 03:59:05 UTC:
    OK. I also have several doubts on the lossless bitstream part of WebP.
    That shall be later posted in my GitHub "talk".

    To my understanding, WebP is composed of 2 separate parts:
    Lossy: Very much mere VP8
    Lossless: Sort of VP9 derived


    "The thread is very old."
    It seems most relevant however.
    Pointless new topics only cause bad aggregation. (various testing has proven)

    I cannot yet directly post there.
    So probably need your help to post the link instead...

    1 more question:
    Can I publish the content of this email thread?
    As which shall aid the general knowledge of WebP.
[ ^
    Do I need to mask your email too?
    Which regardless public in the RFC link:
    https://datatracker.ietf.org/doc/draft-zern-webp/ ] ]

----

[ jzern@google.com @ CE 2024-11-05 03:15:43 UTC:
    “WebP is composed of 2 separate parts:
    Lossy: Very much mere VP8”
    Yes, lossy is the Intra (key) frame from VP8.

    "Lossless: Sort of VP9 derived"
    Not at all. Lossless is its own bitstream:
    https://www.ietf.org/archive/id/draft-zern-webp-15.html#name-specification-for-webp-loss
    ; drawing more from traditional lossless encoding:
    Like Huffman (prefix coding), LZ77, color cache/palette etc.
    .
    VP9 (and AV1) lossless are limited to a 4x4 Walsh-Hadamard transform:
    Which doesn't perform well with generated content, for instance.


    This is fine to post.
    I would say the same if the thread was on "[email protected]". ]

----

[ honorificabilitudinitatibus@tutanota.com @ CE 2024-11-05 19:21:47 UTC:
    OK. Thank you.

    I thought WebP Lossless might be VP9 related primarily for:
    https://www.webmproject.org/docs/encoder-parameters/#vp9-specific-options ("lossless")
    ; and the relationship between VP8.
    It appears it's just codenamed "VP8L" but has few to do with VP8?
    (probably even less connection than JavaScript vs. Java..?)

    My recent testing on lossless VP9 also indicated so. ]

----

[ jzern@google.com @ CE 2024-11-05 21:15:07 UTC:
    Yes, the chunk name, "VP8L" was chosen:
    Mostly for the association to the WebP format, and to avoid any potential naming conflicts in RIFF.
    They are otherwise unrelated. ]
]]