Results 1 to 4 of 4

Thread: jpegraw

  1. #1
    Member
    Join Date
    Mar 2016
    Location
    Croatia
    Posts
    191
    Thanks
    82
    Thanked 13 Times in 12 Posts

    Exclamation jpegraw

    New tool from Cédric Louvrier:

    jpegraw, a fast JPG optimizer

    jpegraw is an experimental, fast JPG optimizer with visually lossless or lossy compression. it uses a recursive multi-processing system

    website, tests: https://css-ig.net/articles/guetzli#test1

  2. #2
    Member
    Join Date
    Mar 2016
    Location
    USA
    Posts
    58
    Thanks
    8
    Thanked 24 Times in 16 Posts
    Targeting DSSIM though. Seems Google has a lot of work ahead to explain the benefits of butteraugli.

  3. Thanks:

    Jyrki Alakuijala (2nd April 2017)

  4. #3
    Member
    Join Date
    May 2008
    Location
    England
    Posts
    325
    Thanks
    18
    Thanked 6 Times in 5 Posts
    These tools(and pingo) are just mozjpeg/zopfli with a different frontend. He's just streamlined the end user experience and added alternative quantisation tables.

    But yeah Guetzli, wow i found this interesting comment on it from one of the contributors:

    https://github.com/google/guetzli/is...ment-276295265

    "Guetzli is the worst on SSIM and PSNR, but best on butteraugli scoring. This is really not a miracle nor a proof of its superiority since guetzli is just a complex optimizer for butteraugli rating."

    "Note, that butteraugli only looks at the worst area of the image, where as ssim and psnr aggregate errors from everywhere in the image."
    Last edited by Intrinsic; 4th May 2017 at 23:03.

  5. #4
    Member
    Join Date
    Apr 2012
    Location
    Stuttgart
    Posts
    448
    Thanks
    1
    Thanked 101 Times in 61 Posts
    Quote Originally Posted by Intrinsic View Post
    "Guetzli is the worst on SSIM and PSNR, but best on butteraugli scoring. This is really not a miracle nor a proof of its superiority since guetzli is just a complex optimizer for butteraugli rating." "Note, that butteraugli only looks at the worst area of the image, where as ssim and psnr aggregate errors from everywhere in the image."
    This is a problem I pointed the Guetzli people at already a year ago (see the separate thread). It is a bad idea to depend on a single image quality index alone, and just optimize to that as objective function. It might look good in its scores, but it might still not look good for the observer. For any serious objective evaluation, I would *at least* include VDP as well and check what it has to say about quality. Then make subjective assessment. Years ago, a student of mine and myself built an MS-SSIM optimized JPEG 2000, just by understanding how MSSSIM works and buildling the scales and pooling into the rate allocator of the EBCOT. It worked great. MSSIM scores of this encoder were way, way, way better than that of the regular JPEG 2000 encoder. Did the images look better? Not really. They looked "different". Artifacts had a different nature, and appeared in different regions of the image compared to the regular JPEG 2000. Conclusion: Don't trust a single metric. Trust your eyes.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •