Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
XYB JPEG: Perceptual Color Encoding Tested (giannirosato.com)
49 points by computerbuster on July 20, 2023 | hide | past | favorite | 44 comments


The clear winner here is JPEG XL, which is fully supported in iOS 17 and will have a billion+ capable clients by end of year.

JPEG XL has HDR, is the best codec for photographic type content in all but extremely low file size, is the best codec for cartoon/chart style graphics, has a container that is enormously flexible (want to store a channel for depth or for object detection? You can do it).


While JPEG XL is great in most aspects, in terms of file size AVIF is substantially smaller at medium image quality. AVIF has a lot of problems though, e.g. no progressive image loading or limited resolution. The situation is really frustrating, since there is only a once-in-multiple-decades chance to replace JPEG and PNG, and this chance should better be met with a format that is great at everything. Because if it succeeds, we will be stuck with its shortcomings for a long time, just as we were stuck with JPEG for decades.


> a format that is great at everything

I too want a magical flying unicorn, and maybe we can make one in the next 40 years, but in the mean time I will happily accept to be stuck with a magical jpegxl unicorn that can’t fly.


You get your unicorn only if it can fly. Otherwise the Chrome team gives you a flying donkey. A non-flying unicorn happens to be not on the table.


i think png will never be replaced because you will always be able to tell that a png file is lossless, but jpeg is almost always used as a lossy format, and you will never be able to tell at a glance whether someone cared enough to make it lossless for a particular file

(note: losslessly preserving jpeg artifacts is still lossless if those artifacts were already present in the source material)


That's a good point, though image viewers could add that information to a "more info" dialogue.


i don't look at dialogs before i download, save, or pick a file from a directory listing


But you look at your image viewer that could have the lossless indicator? (and there is an issue open to add this indicator to the jxl files)

https://github.com/libjxl/libjxl/issues/432


i doubt windows explorer would add an indicator over every thumbnail in a directory listing, and that is a very specific use case (you having a bunch of photos already on your computer and wanting to know if they are lossless), it's incredibly contrived and useless in the real world


Why are you calling your use case contrived and useless? Also, why is windows explorer relevant when there are much better file/image "listers" out there?


it's useless as a solution to the actual problem because usually the filename and an image preview are the only pieces of information you get about an image no matter where you are and saying "well just display the quality level" is incredibly ignorant of the literal entire world not doing this

uploaded a file to discord? preview and filename

found an image on google? preview and filename

opening a file on your computer? i fail to understand how windows explorer is not relevant when you can't just set the image viewer for save/open dialogs on windows, that's just not a thing that's possible to do

and even then, do i want to know the quality level of an image only after saving it manually? also no

tl;dr when you see a .jxl file you have no idea how compressed it is, but when you see .png you know that it's always lossless. it's not hard to understand how png will stay relevant for at least this reason


> usually the filename and an image preview are the only pieces of information you get about an image no matter where you are

that's obviously false, it entirely depends on "where you are"

> ignorant of the literal entire world not doing this

The entire world is ignorant enough to not know that png is lossless

Also, don't hide behind the entire world, this conversation is about you "i don't look at dialogs before i download, save, or pick a file from a directory listing"

> fail to understand how windows explorer is not relevant when you can't just set the image viewer for save/open dialogs on windows, that's just not a thing that's possible to do

(by the way, it'd be entierly possible to add metadata even in Windows Explorer when this https://github.com/saschanaz/jxl-winthumb/issues/26 is added)

but you're not forced to just the save/open dialogs of the default file manager. You can even use a more convenient workflow of navigating in your file manager and making the open dialog switch there, and within your file manager you would have more info vs WinExplorer The "save" dialog is worse, so that would require implementation of the feature mentioned above

> when you see .png you know that it's always lossless.

that's also obviously false, lossless is a feature of the workflow (your note conveys some familiarity with this simple fact), not just the file format, and since you don't know how much loss there was before you get to download the final PNG, you can't say that it's lossless Yes, that's a useful heuristic, but then it's just that, not "always"

Likewise > and even then, do i want to know the quality level of an image only after saving it manually? also no

For the same reason, you actually don't know the quality level of an image for any format

> it's not hard to understand how png will stay relevant for at least this reason

it't not hard to understand, yet you don't. PNG will stay relevant simply because widespread formats are sticky and it takes a while to switch even to a strictly superior alternative like JXL


> that's obviously false, it entirely depends on "where you are"

i said "usually"

> > it's not hard to understand how png will stay relevant for at least this reason

> it't not hard to understand, yet you don't.

i said "at least"

this argument isn't productive anymore and i won't be continuing it, have a nice day


> i said "usually"

still false since in addition to the dumbed down web platforms you've included a PC in the set And your example is wrong - just checked that google images doesn't show an extension (the link is too long to see it, and some links don't even have it, and in a mac save as dialog you also don't see it), so PNG imaginary advantage isn't even "usually" there

> i said "at least"

this also doesn't help you since your reason is incorrect, so it can't be a factor

> this argument isn't productive anymore and i won't be continuing it, have a nice day

lol, you've missed the chance to make it productive by ignoring the core issue that PNG doesn't signal losslessless


But you can't tell that png is lossless as you don't know how much loss there was before png


Confirming that JPEG XL is a winner. Here's one of the main developers of the format, Jon Sneyers, writing 3 days ago:

JPEG XL: How It Started, How It’s Going

https://cloudinary.com/blog/jpeg-xl-how-it-started-how-its-g...


The clear decade-plus-of-backwards-compatibility-in-globally-deployed-software winner is XYB JPEG, a category in which neither XL not AVIF are valid comparisons at all.

This in no way contradicts your points about the forward opportunities of XL. For anyone that seeks the best backwards-compatibility for display use, there’s now XYB as well.


In the article they mention JPEG XL (aka .jxl) format that I hope everyone knows about:

https://jpegxl.info/

It is supported by Adobe, included in Safari 17, is behind a flag in Firefox, and was behind a flag in Chrome too until they idiotically chose to remove it despite severe pushback from the community. My take is that it will be one of the most-popular image formats in the coming decade.


I hope I can start saving game screenshots in JXL soon. Pretty sure the HDR is lost when saving as jpeg.


> Pretty sure the HDR is lost when saving as jpeg.

Nvidia can capture HDR screenshots in JXR format in supported games.


given Chrome's stranglehold how do you see it becoming one of the most popular image formats in the coming decade.


Their decision is schizophrenic: the JPEG XL format is a combination of two technologies: PIK (developed by Google) and FLIF (developed by Cloudinary) - one of them developed by Google. It is entirely sensible for them to reverse their decision, which came before Safari announced the format would be supported.

Adobe is already supporting the format, and many image editors / browsers are including it.

Finally, adding a WASM polyfil to your website would still yield many benefits, enough to outweigh not-100%-yet-support-by-browsers.


Not schizophrenic, it was political.

The guy behind web-p has more clout with the chrome team.


From my very very limited understanding, the amount of influence webp (compared to even say AVIF) people have seems disproportionate compared to how useful it's been.


wdym webp is really good at serving artifacted images with slightly fewer bits per pixel than jpeg. nobody uses jpeg qualities over 50, right?


From the same author: https://giannirosato.com/blog/post/image-comparison/

"Considering JXL has been endorsed by Facebook, Adobe, Intel and the Video Electronics Standards Association, The Guardian, Flickr and SmugMug, Shopify, the Krita Foundation, Serif Ltd, Gaia Sky, and many more, the market is most certainly interested. My current optimistic hope is that JXL takes off outside the web among professionals working with tools like the Adobe suite or alternatives, and camera manufacturers, smartphone OEMS, and others take notice & begin to think about JXL more seriously. The benefits cannot be ignored, and it is (in my opinion) the only image format that is in every way superior to JPEG & offers a concrete future for the many existing JPEGs on the Web & beyond."


Decisions lightly made can be lightly reversed. It's not exactly hell freezes over situation to see jxl reintroduced to Chrome


h.264 did pretty well even if Chrome decided not to support it


Because it was used in Blu-ray discs, streaming apps, and TV. But JPEG and any successor is mostly a format for the Web and thus web browsers.


The situation was a bit different, as (IIRC) h.264 already had an enormous amount of usage.

JXL is basically starting from zero.


That is crazy indeed.

I can't remember, do the other encoders automatically change the chroma subsampling? XYB JPEG peaks pretty hard at specific bpp ranges depending on the subsampling.

Also, things like this make me wonder what other tasks would be better in XYB colorspace. BM3D, for instance, performs best in "opponent" color space: https://github.com/HomeOfVapourSynthEvolution/VapourSynth-BM...


mozjpeg automatically handles chroma subsampling, scaling up from 4:2:0 to 4:2:2 & then 4:4:4. WebP is 4:2:0 only, AVIF is 4:4:4 only (as tested) & I believe JXL doesn't subsample at all.


wonder if there is a perceptual color system that accounts for the rare fourth color receptor...

https://en.wikipedia.org/wiki/Tetrachromacy#Tetrachromacy_in...

don't know if that would be reproducible on a regular RGB display, though


> don't know if that would be reproducible on a regular RGB display, though

It wouldn't, and also not with standard camera sensors. The RGB subpixels correspond to the main wave length sensitivity of the three color cells in the eye. I think for tetrachromy there is a fourth cell for UV light with shorter wavelength than blue. There is no way to support this with our current tech stack.


this reminds me of how designing for "the average human" leaves out anyone with even slightly superhuman capabilities, and how frustrating this is for us when our capabilities surpass what designers thought were good enough. for us, this happens frequently with monitor PPI and HMD PPD, but i can imagine someone with tetrachromy theoretically being able to perceive colors that computers can't reproduce, and wondering why computers lack these colors, only to find that this is supposedly an impossibility that was optimized out

as an example, apple's new HMD advertises things that simply can't possibly be true, so we probably won't be getting one any time soon. there's zero reason to upgrade from a simple hp reverb g2 when the only difference between two blurry headsets is that one of them costs four times more. we'd rather be conscious of this than try to fix it when no solution currently exists

this isn't saying displays for tetrachromy should have been the norm, as that's very wasteful on most people. but maybe displays for tetrachromy could have existed, somewhere, in some market, just like tech devices with placebo-level specs tend to exist in some market


I was surprised by your pronoun usage, then I clicked on your profile. Reminded me to think of the ambiguity of the English "you", in contrast to "I"/"we".

Edit: To add a bit to the topic. Tetrachromy seems a quite interesting example for "superhuman" sensory ability. Normal people have three cone cells, and the maximum/minimum activation of each cell gives rise to 2^3=8 "primary" subjectively dissimilar colors: 000 (black), 001 (blue), 010 (green), 011 (cyan), 100 (red), 101 (magenta), 110 (yellow), 111 (white). All the others look like blends of the former. But for tetrachromy there are 2^4=16 different such "primary" colors. That's a substantial difference.


yeah i wish there were an easier way to say we have DID than making multiple HN accounts, we were even personally recommended to do that by HN's moderation team but opted not to because we need the shared rep and comments.

including a disclaimer in every comment (which we have to do if we include a signature, because then people will ask what the signature means) is also somewhat tiring, so right now the pronoun usage is a bit of an unsolved problem, hehe.

glad to see the profile did its job though

as for the edit, another commenter says there's still no evidence that people with true tetrachromy have a fourth primary color. that's not to say they don't, but just that it may be years or decades until a study conclusively proves its existence, if there is one


TLDR: A display for people or other creatures with true tetrachromatic vision (not just an additional receptor, but the ability to perceive it as a distinct primary colour) would need to be able to output that fourth primary colour, but there doesn't seem to be enough data to determine conclusively at this time if women with a fourth type of cone perceive a fourth primary colour.

The one study I've been able to read the full text for that involved determining spectral sensitivity of a woman with distinctly tetrachromatic vision[1] indicates that she had some cones with a peak in-between what most people would see as red and green,[2] and could consistently and accurately discriminate between colours that would appear identical to most people.[3] Unfortunately (and surprisingly, IMO), it doesn't discuss her subjective perception, i.e. does she perceive a true fourth primary colour?

It seems like a very, very rare gift, so IMO it's hard to even determine if the women who have it respond in similar ways, which would be a prerequisite for developing a display for them. Everyone else the authors tested who had a fourth type of cone couldn't actually discriminate between colours beyond someone with the typical three types. It would be interesting to see if they gained the discrimination ability by wearing glasses with extremely narrow bandpass filters for their cones' center wavelengths, like the ones that give men with red/green colourblindness the ability to distinguish between red and green.

[1] https://jov.arvojournals.org/article.aspx?articleid=2191517 - which cites a much earlier paper by two of the same authors (https://pubmed.ncbi.nlm.nih.gov/8351822/) that also included the subject "cDa29". [2] Search for "Simulated cone fundamentals for cDa29" in the first paper linked in [1] [3] "one participant cDa29 [...] behaves as if she has access to an additional cone signal: she made no errors at any value of (R670 / (R670 + G546)) [...] and her response times not only are faster than those of other participants [...] but are roughly even across the stimulus space"


thank you <3


There isn't even a way to account for individual differences, let alone for things like tetrachromacy. Color blindness is probably easier (at least crudely), but AFAIK it also varies between the same types. So unless you profile your perception individually, we're stuck with the biggest cluster of the "average human vision system".


Reading how tuned this is for the quirks of human perception, it makes me wonder how difficult alien technology would be for us to comprehend.


i see different colors per eye so yeah


Images are completely green in Safari 15 and Photoshop. Works fine in Chrome and Firefox


How does the new cjpegli encoder using the XYB color space stack up compared to the competition?




Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: