Google frees nifty ML image-compression model... but it's for JPEG-XL A new application of machine learning looks both clever and handy, as opposed to the more normal properties of being somewhere between privacy-, copyright-, or life-endangering. But before you get too excited, you can't have it. The true cost of ML applications varies. Many are free to use, which means they endanger the paid …
You have two choices: "Fuck the future so we can have it large" and "do the right thing".
Then there are varying degrees between. The question is how much do you want to do the right thing and how much do you want to screw over your future self and your children?
If your politics favours spending your hypothetical children's hypothetical college fund on a trip to Vegas, then have a great trip.
El Reg lost a hell of a lot more when they opened up the editorial to the none UK litteratii.
No slurs intended peeps, English is English, as fucked up as the language is, it is unique for opportuning such delights as "humour" out of these ambiguities, ie; irony for a start.
What we in UK have enjoyed from this greater diversity of tech input has unfortunately, watered down our previous sense of enjoyment, having been created by such a warped sense of humour which often lays bare the marketing wallabies best efforts to confuse and bamboozle us. Dabbsie etc.
As I haven't been there it is probably incorrect for me to have an opinion on our friendly US and Oztrralian scribes, but from reading the lines here in el Reg, (as there are now less lines in between to read) it appears that OZtralian and US scribes cannot tell the difference between talk and mutter........ Having attempted to read the "news" on BBC Home recently, I assume they have taken a similar globalizing route to publishing news about the news, but not the "actual news" you thought you were gonna get......
Alas, our loss is also to our benefit..........?
ALF
> the horrific loss of ice-moving jobs when the freezer was invented?
Supplying ice to the eastern US was a BIG industry in Maine for a few decades. Towns on ice-ponds prospered. Major ice-houses were built by the railroads and harbors. Men and boys sawed ice (engine saws were a very late development).
"At its heyday in 1870-1890, some 25,000 men converged on the Kennebec ice fields each winter to cut and store ice. ...In these decades Maine’s ice returned a wealth greater than that of California’s annual gold production." cite
The idea is that a low-res version of the whole image appears right at the start, and by the time that your visual cortex has decided where to point your pupils, that area of the image is already getting sharpened up.
Can you please tell this to the web designers and coders who insist on requiring javascript to see the images? Many sites simply don't display any image without scripts and a lot do a very poor job at reimplementation of async loading which results in forced download of the whole image before putting it into the DOM for rendering.
So, in theory it is nice to have progressive images was it not for (very poor) designers and (bad) front-end coders to fuck things up.
/rant
Ah, is that what it's supposed to be? Thanks for the info.
Right now I'm on a 300Mbps connection. It's OK.
At other parts of the year I have to make do with 10Mbps. Strangely enough that's OK too. I have a strong suspicion that web page rendering speed is, in 2022, much more dependent on how fast your browser can process JavaScript code than on how fast it can get bytes off the internet.
I'd like to get in right now by stating that this is not the fault of JavaScript. Slow crap code does not require any particular language.
-A.
It's been observed before but the quick and easy way to both speed rendering of a page and delivery thereof is to kill all the adverts and most of the scripts... though two of the 'grey blob placeholder' culprits are Youtube and Paypal.
I'm sure that when the internet was tiny that jpegs were delivered in progressive mode.
It's extremely easy to progressively load in scripts, webpack supports it natively. I don't understand why everyone isn't doing it.
Additionally, there are a lot of reasons many sites are going JS only. Two big ones are that it reduces the total data transfer (if done right), with the scripts coming off a cheap and fast CDN instead of an app server. And the second being that you can do things to support complex UI interactions that would be difficult or inefficient with server side rendering.
P.S. I also hate the "grey blobs", which the UI people call skeletons, they take longer to load than the real content most of the time
In our race to automate (improve ?) everything - we the able bodied often forget to include the likes of folks who are less well endowed with stuff like, vision - for a start.
In my own ADHD world, it appears that it has been decided by none ADHD people that ADHD people do not focus on the same bits of an image, or sound, or vocabulary that the other, none ADHD people do. Er, aren't these the people who are designing interrfaces for us ADHDers to use.............
When there are so many problems that need solving I am constantly amazed at the number of problems that are being invented currently which disbaled and able bodied people are supposed to solve before they can get any fucking work done.....
Is this more AI ? Artificial Idiocy at work - as usual I say, yawn. The human brain !
ALF
But JPEG already supports bandwidth-limited loading, the Progessive save pattern. 3 or 4 passes to decompress, allowing early low-res loading, progressing to high resolution through time.
So why are we so worried about recreating the wheel? Just allow & use progressive JPEG and be done with it.
> looks like a spectacularly inappropriate use of AI.
That seems like a bit of an over reaction. Google haven't enslaved a General Systems Vehicle's Mind, forcing it to look at FaceBook image posts all day.
They are just running a lump of code to do a job: a trained tensorflow-lite model which decides which bit of the image to put out in full detail next. Really, it is doing little more than the existing progressive scan does, except instead of having a "top to bottom (skipping k rows), left to right (skipping j columns), repeat with offsets until the whole image is done" co-ordinate generator it has a "pick an interesting bit, pick another until the hole image is done" generator.
So, do you object to using that many CPU cycles "just to compress an image"? Many cycles are used just to generate stuff that someone hopes is pleasing to the eye.
Or do you find the idea of training a ML model to this task inappropriate? It probably kept someone gainfully employed for awhile and one can think of other uses for this or a similar model - how about setting camera focus? Always worth at least trying an idea...
Satisfy my curiosity - what is so 'inappropriate" here?
What about first stripping of the metadata crap that seems to grow wild on pics these days?
I'm always amazed if I get a pic that came out of e.g. PS or a phone being several MB, then stripping of the metadata which reduces its size to the kB ranges.
And then also ask yourself which and/ or which user NEEDS that metadata...
> And then also ask yourself which and/ or which user NEEDS that metadata...
That much metadata is intended not for the Users.
Now, be a good boy and just put the picture from your 'phone straight, they'll know what to do with it.
Remember, render unto Meta that which is meta.
Barnsley's fractal image compression was very impressive in Microsoft's Encarta, and then it, along with Irritating Systems seemed to disappear - I recall using:
https://en.wikipedia.org/wiki/Genuine_Fractals
...it was very useful actually. I wish there was a plugin for video which does the same.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
