Google's Jpegli Offers ~35% Compression Improvement For High Quality JPEGs

[SciK]

now the colorspace bit is a wee bit different as that relies on ICCs and there are plenty of image software that doesn't work with ICC, but thats not really anything new. Jpegs with ICCs already don't display properly in these image viewers, but now it will look more broken. IIRC this is optional? I can't remeber

Optional and off by default in cjpegli. You need --xyb to use it.

[risho]

what are the tradeoffs between --xyb and not using it? does it improve the compression ratio? and when you say it won't display properly in some image viewers how improperly? does it just look a little weird or is it completely broken? how frequent will you run into this incompatibility? just if you are using extremely old software from the mid 2000s, or is it common even with image viewers of today?

[Quackdoc]

what are the tradeoffs between --xyb and not using it? does it improve the compression ratio? and when you say it won't display properly in some image viewers how improperly? does it just look a little weird or is it completely broken? how frequent will you run into this incompatibility? just if you are using extremely old software from the mid 2000s, or is it common even with image viewers of today?

xyb will encode the image using the XYB colorspace, This allows a some neat features when it comes to compression and presentation. the big benefit is allotting bits to colorspaces in accordance with how humans actually see colour, since XYB is a colorspace based on how humans perceive color (LMS) and not how displays emit color (RGB). This gives us a YUV422-esq bit depth savings while being perceptually much higher quality. It also allows a much larger color range then gamuts like sRGB, it can also maintain color accuracy when displaying across color gamuts, though gamut mapping can still bite you in the arse.

as for how broken it is, a lot of software supports ICC profiles but some don't still. It's kinda hit or miss afaict, ffmpeg based image viewers often don't support ICC for a myriad of reasons. Qt software iirc should work fine.

EDIT: attached is a png with XYB profile from libplacebo gitlab issue tracker so you can test it in your own viewers, it will be very apparent if it doesn't work, gwneview and loupe work, koko doesn't if you want known good and bad viewers https://0x0.st/H8aK.png

[Radtraveller]

Yeah.. google “declined” jpegxl.. because they were going to add a few lines of code and use jpegxl to create their own “jpegli”.. sounds about right for them.

[Old Grouch]

Remember, this is google we're talking about! They're into R&D of ADVANCED computing technologies AT SCALE.
So they just use R&D approaches that can SCALE to handle NP-hard problems, WHATEVER the cost.

They started out with a roadmap to ramp their literal(!) room full of monkeys with keyboards to 1-google (the number) primate-bits scale.
Sure there's no actual uniform direction of what their individual monkey-developers are working on, that's the point! --
They could repeat themselves, duplicate effort, whatever, but they're just as likely to randomly go and try something truly unexpected and groundbreaking!

Not only will they eventually find A GOOD solution, they'll find EVERY good (and bad!) solution, and in the mean time we'll be unwilling unwitting alpha-testers of them all!

Then as technology advanced they've started embracing quantum computing and now have data centers full of quantum-nano-monkeys working alongside maxwell's daemons!

Again, yes, it seems like they're working on any given problem in all different for-x, against-x, sometimes-x, somewhat-x directions, but that's the beauty of the wavefunction! When they can finally get to trying to solve every possible problem in every possible way simultaneously then the wavefunction will collapse and like Dijkstra's shortest path algorithm, the ONE TRUE SOLUTION, you know, to life, the universe, and everything will emanate!

See, that is the beauty of THE GOOGLE!

You are appearing a bit manic there. If it is for effect, fine. If not, you might need to look at your drug dosing.

Edit to add: If this seems a bit odd, I have had to deal with a couple of manic-depressives in my time, one of whom appreciated being told that they were going into their manic phase so they could deal with it effectively. Another did not, and would deliberately omit their medication when getting manic. Which was a shame, and did not end well. I hope, if it applies, that you are the first type.

[Old Grouch]

If you where switching to anything else, that would need to be standardized first, which is a long and tedious process.

Not if you have monopoly power. Then any decision you take becomes the standard.

[pong]

You are appearing a bit manic there. If it is for effect, fine. If not, you might need to look at your drug dosing.

Edit to add: If this seems a bit odd, I have had to deal with a couple of manic-depressives in my time, one of whom appreciated being told that they were going into their manic phase so they could deal with it effectively. Another did not, and would deliberately omit their medication when getting manic. Which was a shame, and did not end well. I hope, if it applies, that you are the first type.

It was just supposed to be CS-google-humor-ish.
If I had XKCD-art-talent I suppose it would have come off better that way.
But it's good you're trying to be helpful, as you said, many need it.

[Quackdoc]

Yeah.. google “declined” jpegxl.. because they were going to add a few lines of code and use jpegxl to create their own “jpegli”.. sounds about right for them.

people keep posting this, and it keeps being wrong. jpegli is part of jpeg-xl repo, and is developed by the same people, a few of the important ones being google employees

[Old Grouch]

It was just supposed to be CS-google-humor-ish.
If I had XKCD-art-talent I suppose it would have come off better that way.
But it's good you're trying to be helpful, as you said, many need it.

I'm glad you are OK.

And sorry for using the term manic-depressive. I should probably have said bipolar, or whatever the current term is. The 'Old' in 'Old Grouch' applies.

[Artim]

Not if you have monopoly power. Then any decision you take becomes the standard.

For that you'd need to have that power. Google does have that power, still everything they do is going through the long standardization processes. And by no means that is just getting accepted. WebGPU for example could have been just a web version of Vulkan. But because Microsoft and Apple needed to have a say in it, it had to be the lowest common denominator between Vulkan, DX12 and Metal. And lowest common denominators are always the worst compromise. Same with QUICK. It was a great idea bringing a massive security boost, yet it was extremely water