JPEG Committee is Banking on AI To Build Its Next Image Codec (thenextweb.com) 43
Joint Photographic Experts Group (JPEG), a committee that maintains various JPEG image-related standards, has started exploring a way to involve AI to build a new compression standard. From a report: In a recent meeting held in Sydney, the group released a call for evidence to explore AI-based methods to find a new image compression codec. The program, aptly named JPEG AI, was launched last year; with a special group to study neural-network-based image codecs. Under the program, it aims to find possible solutions towards finding a new standard. To do that, it has partnered with IEEE (Institute of Electrical and Electronics Engineers) to call for papers under the Learning-based Image Coding Challenge. These papers will be presented at the International Conference of Image Processing (ICIP) scheduled to be held at Abu Dhabi in October.
Smart (Score:5, Funny)
They are running it against images gathered on the Internet. The result is the new codec will be able to compress images of cats and naked women by 99%.
So their new standard will be the ... (Score:5, Funny)
Joint Pussy Expert Graphics compression standard. :)
They also hired... (Score:2)
They also hired some big names in AI including Al Gorithm and Sandeep Lehrning, The committee also hired Erma H. Gard as a meme consultant.
Re: (Score:2)
Do we need a new codec? (Score:2)
While I am all for efficiency and using as little bandwidth as possible. Being the standard home network connections are now measured in gigabits per second or more commonly in megabits per second vs kilobits per second, or bits per second which were common 20/30 years ago. Most speed inefficiencies are not from our image codecs but from people not using the existing ones, Not saving images in the appropriate size, and most commonly hooking over to external (ad servers) which just take so long to respond
Re:Do we need a new codec? (Score:5, Informative)
Re: (Score:2)
Re: (Score:2)
More people are accessing the internet via phones than computers on a home ISP connection. And many of those people are still paying by the GB of data transferred. More compression, less data sent.
So maybe we should be looking at a way to "compress" the several megabytes of Javascript pulled down by every page for tracking purposes.
All I'm insinuating is that in 2005, "the problem" was that image files took up huge amounts of a web page's bandwidth use. Today... it's crap we didn't ask for. Yes, I recognize that improving two things is better than improving one thing. But personally I'd rather pay more monthly to transfer inefficiently compressed content over all the advertisement and metrics-ga
Re: (Score:2)
We desperately need an archive format more (Score:2)
So what does one do? Photo tools seem to like TIFF...because apparently disk space means nothing to people....PNG seems good, but no one rea
Re: (Score:2)
What you're asking for is pretty much impossible. RAW files are the most storage efficient because they accurately reflect the actual amount of data acquired by the camera sensor (and are usually losslessly zipped on top). When you convert a RAW to TIFF it at least triples in size because instead of having to store only one value per pixel like the RAW, now it has to store one each for the R,G and B channels. You can losslessly compress that (TIF usually does) but it's still bigger.
If you want lossy compres
Re: (Score:2)
RAWs straight out of camera are actually already compressed. This is why the file size varies from image to image of the same exact resolution. This is also why file sizes are larger for higher ISO (more grain, less commonality within the pixels). If you want the fastest way to check this, just look at your camera's "remaining photo" counter (whatever it happens to be called), and adjust from ISO 100 to ISO 6400. Without taking a single pic, the "available" storage will appear to drop. This is also why when
Re: We desperately need an archive format more (Score:2)
Not true. There are hundreds of RAW formats and most are not compressed. Many are variations on TIFF type.
Re: (Score:2)
"File compression techniques that lose some of the original data are called ‘lossy’. However, ‘lossless’ compression is also available. This is used by Canon digital cameras when the highest resolution image is stored as RAW data."
Literally Canon's own documentation. And yes, there are other formats out there by other manufacturers, but they're similar nature.
https://cpn.canon-europe.com/c... [canon-europe.com]
Re: (Score:2)
New iPhones and some high end cameras also support a format called HEIC (high efficiency image codec) which I understand involves using x.265 video compression on a single (presumably huge) bitmap.
This doesn't help RAW or DNG in particular, but I understand it to be a less lossy yet smaller output format when compared to JPEG, albeit at a cost of greatly increasing the CPU/GPU needed to decode it.
I pulled some HEIC files off a dead iphone a couple weeks ago. I'd never even heard of the format before. It wou
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Being the standard home network connections are now measured in gigabits per second or more commonly in megabits per second
There are still tons of real-world cases to be found using cellular data, where you are hard pressed to get even 1MB/s. Compression still matters very much.
Re: (Score:2)
YES, bandwidth still matters.
NO, we don't need yet-another-codec. We already have FLIF [flif.info] (Free Lossless Image Format) for BOTH lossless and lossy [flif.info] encoding.
Yet another image format (Score:4)
Re: (Score:2)
I've still seen more webp images than jpeg 2000, and it's far newer.
Re: (Score:2)
Looks like webp has better support; that likely has a lot to do with it.
webp [caniuse.com]: works in everything except IE and Safari.
jpeg 2000 [caniuse.com]: only works in Safari.
Re: (Score:2)
I'd bet the most common use of jpeg2000 is in PDF files, but Acrobat 5 is still a pretty common target compatibility and doesn't support it, and even the Adobe presets for newer versions still default to regular jpeg.
Acrobat 6+ support jpeg2000, so that's a lot of instances of software.
Re: (Score:2)
The most common use of JPEG2000 is digital cinema. Every frame in a Digital Cinema Package (DCP) is JPEG2000, and it's heavily compressed.
Commercial editing software outputs direct to JPEG2000 and thence to DCP, but I did it once on Debian using ffmpeg and a toolset called OpenDCP.
The original MP4 was exported to a directory of frames in lossless TIFF - each frame ~2 - 4 MB, then those were converted to JPEG2000, where each frame was about 1/10 the size of the TIFF. And no compression artifacts even when sh
Re: Yet another image format (Score:2)
Since it allows for frame exact splicing and frame by frame editing with minimal loss.
Also, since the format can be lossless, one could presumably do the editing lossless if needed.
Re: (Score:2)
It's possible, but it's more common to do much of the editing on a low-res proxy. That way you can use commodity PCs for a lot of the editing, and only bring in the heavy-hitters when needed, e.g. compositing high-def CGI into the "live" footage. I had to edit a low-def copy of my first short film, because the footage from the camera was 1280x720, and my computer at the time couldn't handle that. Edit the proxy at leisure, then apply the editing cuts & transitions to the high-def footage for final outpu
Re: (Score:2)
Re: (Score:2)
I had two yesterday here (https://en.todocoleccion.net/postcards-religious-in-memoriam-cards/bonita-estampa-religiosa-antigua-sanctus-rochus~x41275298)
PDF supports jpeg2000, but I don't think it's very often used.
Re: (Score:2)
I had two yesterday here (https://en.todocoleccion.net/postcards-religious-in-memoriam-cards/bonita-estampa-religiosa-antigua-sanctus-rochus~x41275298)
PDF supports jpeg2000, but I don't think it's very often used.
The main image on that page looks like a jpg to me.
Re: (Score:2)
So it does.
right click save as gets a webp though
Re: (Score:2)
<picture>
<!--[if IE 9]><video style="display: none"><![endif]-->
<source type="image/webp" srcset="https://cloud10.todocoleccion.online/postales-religiosas-recordatorios/tc/2014/01/28/00/41275298.webp">
<!--[if IE 9]></video><![endif]-->
<img src="https://cloud10.todocoleccion.online/postales-religiosas-recordatorios/tc/2014/01/28/00/41275298.jpg" alt="Postales: BONITA ESTAMPA RELIGIOSA ANTI
Re: (Score:2)
JPEG2000 is used when you need lossless encoding. It's used by some scanning machines that can directly send you an email with the scanned document as an image. The document is either in PDF or JPEG2000 format.
Re: (Score:2)
I've still seen more webp images than jpeg 2000, and it's far newer.
If I recall correctly there was a lot of fear that wavelet compression was covered by patents and nobody big wanted to take point on it. Whereas webp is just the single image version of webm, which Google is using extensively for YouTube. So it should be very safe from an IP rights perspective.
I see FOUR DOTS! (Score:2)
Count the dots.
- Dots?
The dots! Look, look.
"Every single detail is", dot-dot-dot, "perfect".
Okay? There are three dots in that ellipsis, which is the correct number of dots.
However, in the message that I sent her, there are four dots, which is incorrect.
Thus, not perfect.
It was a joke.
- Was it?
'Dictionary Lookup' for images. (Score:2)
The best text compression algorithms use dictionaries common to both the encode and decode ends, I guess this is the image-based equivalent. Need the AI to do the searching for optional representations within the scope of the dictionary.
Blockchain (Score:4, Funny)
They should definitely use blockchain in the new format. AI just isn't enough.
Translation: We're too lazy, to come up with a goo (Score:2)
...so we're gonna use a universal function and gradually alter its weights until we think it does what we want.
Because Oooohh, "AI"!
This is literally the same as if some company in the 50s had gone "Try our new WonderJelly! Just as tasty as the old one. But now with more *atom*!"
I guess they haven't heard of EXTRA BIG ASS FRIES yet ...
Will it be able to determine SEX? (Score:1)
Ahem, it's "sex" not "gender"; the latter applies to words, not objects.
https://tech.slashdot.org/stor... [slashdot.org] seems to indicate we're moving in the wrong direction.
Vulnerability checking? (Score:2)
How is anyone gonna check AI-written code for vulnerabilities?
Will there even BE human comprehensible source code produced?
I suppose an AI-BlackHat might work, but that would inevitably escape into the wild.
Already done? (Score:2)