News for hackers. Stuff that matters.
Re:Grand Image Compression Challenge at ICIP 2016 by pornel on Tuesday December 08, 2015 @10:46AM (#10697013)
I hope evaluation will use more diverse set of images than the subset given. Only a couple of example images have highly saturated color, and all example images will tolerate poorly done chroma subsampling.

On the web there are categories of images, such as logos, screenshots, renders, large icons, photos with captions, that aren't simple enough to compress well with lossless encoders, but become a mess when compressed with codecs tuned for high-res photos without any sharp highly saturated edges.

Re:Grand Image Compression Challenge at ICIP 2016 by ygra on Tuesday December 08, 2015 @01:41PM (#10698383)
Well, it is JPEG, after all. So their focus is most likely photographic pictures. As evident by sentences as »contributors to this challenge may want consider contribution of their technology as input to the ongoing JPEG activity«.
Re:Grand Image Compression Challenge at ICIP 2016 by vardump on Tuesday December 08, 2015 @02:58PM (#10698986)
You don't need to subsample chroma in JPEG. I often disable it (use 4:4:4 sampling) if there is high color contrast detail. Say a tree with red fruit, marketplace, kid's toys, etc.
Re:Grand Image Compression Challenge at ICIP 2016 by pornel on Tuesday December 08, 2015 @05:29PM (#10700173)
I know it's not necessary in JPEG, but some codecs (especially video and therefore video-codec-derived still image formats) have only 4:2:0 option. And if a codec chooses to use it, I'd prefer the test suite to require it done well.

I've recently looked at it closely and found that almost every codec does chroma subsampling incorrectly, but the error is visible mostly in computer-generated graphics, and rarely in photos, and probably that's why nobody cared to fix it.

https://github.com/mozilla/mozjpeg/issues/193