MHLoppy

Sorely missing a com/mag/sub/your-preferred-jargon-word for theocho right now

I actually think this video is doing a pretty bad job of summarizing the practical-comparison part of the paper.

If you go here you can get a GitHub link which in turn has a OneDrive link with a dataset of images and textures which they used. (This doesn't include some of the images shown in the paper - not sure why and don't really want to dig into it because spending an hour writing one comment as-is is already a suspicious use of my time.)

Using the example with an explicit file size mentioned in the video which I'll re-encode with Paint.NET trying to match the ~160KB file size:

Original PNG (5,725,123 bytes) https://files.catbox.moe/l2wsxr.png
Optimized PNG (5,579,693 bytes) https://files.catbox.moe/2dh00n.png
Lossless WebP (4,504,946 bytes) https://files.catbox.moe/hiephv.webp
Lossy WebP; default preset, quality 13 (162,820 bytes) https://files.catbox.moe/jnv4tp.webp
JPEG XL; quality 12, effort 7 (164,859 bytes) https://files.catbox.moe/fylcqa.jxl (your browser probably can't view this image)
AVIF; lossless alpha compression [shouldn't matter], quality 44, very slow preset, 4:2:0 chroma subsampling, preserve existing tile size, do not use premultiplied alpha (164,855 bytes) https://files.catbox.moe/u7q7dn.avif

Hadriscus has the right idea suggesting that JPEG is the wrong comparison, but this type of low-detail image at low bit rates is actually where AVIF rather than JPEG XL shines. The latter (for this specific image) looks a lot worse at the above settings, and WebP is generally just worse than AVIF or JPEG XL for compression efficiency since it's much older. This type of image is also where I would guess this type of compression / reconstruction technique also does comparatively well.

But honestly, the technique as described by the paper doesn't seem to be trying to directly compete against JPEG which is another reason I don't like that the video put a spotlight on that comparison; quoting the paper:

We also include JPEG [Wallace 1991] as a conventional baseline for completeness. Since our objective is to represent high-resolution images at ultra-low bitrates, the allow-able memory budget exceeds the range explored by most baselines.

Most image compression formats (with AVIF being a possible exception) aren't tailored for "ultra-low bitrates". Nevertheless, here's another comparison with the flamingo photo in the dataset where I'll try to match the 0.061 bpp low-side bit rate target (if I've got my math right that's 255,860.544 bits):

Original PNG (2,811,804 bytes) https://files.catbox.moe/w72nsv.png
AVIF; as above but quality 30 (31,238 bytes) https://files.catbox.moe/w2k2eo.avif
JPEG XL could not go below ~36KB even at quality 0 when using my available encoder, so I considered it to fail this test
JPEG (including when using MozJPEG, which is generally more efficient than "normal" JPEG) and WebP could only hit the target file size by looking garbage, so I considered them to fail this test out of hand

(Ideally I would now compare this image at some of the other, higher bpp targets but I am le tired.)

It looks like interesting research for low bit rate / low bpp compression techniques and is probably also more exciting for anyone in the "AI compression" scene, but I'm not convinced about "Intel Just Changed Computer Graphics Forever!" as the video title.

As an aside, every image in the supplied dataset looks weird to me (even the ones marked as photos), as though it were AI-generated or AI-enhanced or something - not sure if the authors are trying to pull a fast one or if misuse of generative AI has eroded my ability to discern reality 🤔

edit: to save you from JPEG XL hell, here's the JPEG XL image which you probably can't view, but losslessly re-encoded to a PNG: https://files.catbox.moe/8ar1px.png

Really wish they published the whole dataset. They don't specify on the page or in the paper what the full set was like, and the GitHub repo only has one of the easy-to-read ones. If >=10% of the set is comprised of clock faces designed not to be readable then fair enough.

The human level accuracy is less than 90%!?

The loss brainrot has taken another victim 😔

the kbin.social domain expires in a few days, on 10 September

Looks like someone might've bought the domain?

It doesn't count when you have to change the headline

The Rules
Posts must be:
Links to news stories from...
...credible sources, with...
...their original headlines, that...
...would make people who see the headline think, “That has got to be a story from The Onion, America’s Finest News Source.”

Poor proton-Earth, barely mentioned at all :'(

Anecdotally, quite a lot of users vote "selfishly" and don't care that downvoting reduces visibility. all and local feeds also fall victim to people voting as if these are their own personal curated feeds.

And I hate it 🫠

"We're going to collect as much data about you as we can to sell to advertisers"

That's a rather pessimistic interpretation of a privacy policy that starts with this:

The spirit of the policy remains the same: we aren’t here to exploit you or your info. We just want to bring you great new videos and creators to enjoy, and the systems we build to do that will sometimes require stuff like cookies.

and which in section 10 (Notice for Nevada Residents) says:

We do not "sell" personal information to third parties for monetary consideration [as defined in Nevada law] [...] Nevada law defines "sale" to mean the exchange of certain types of personal information for monetary consideration to another person. We do not currently sell personal information as defined in the Nevada law.

So yes, I suppose they may be selling personal information by some other definition (I don't know the Nevada law in question). But it feels extremely aggressive to label it a "shithole" that "collect[s] as much data about you as we can to sell to advertisers" based on the text of the privacy policy as provided.

I guess perspective here depends on your anchoring point. I'm anchoring mostly on the existing platform (YouTube), and Nebula's policy here looks better (subjectively much better) than what runs as normal in big tech. If your anchor is your local PeerTube instance with a privacy policy that wasn't written by lawyers, I can see how you'd not be a fan.

However beyond being in legalese I'm not sure what part of it you find so bad as to describe it as a shithole. Even compared to e.g., lemmy.world's privacy policy Nebula's looks "good enough" to me. They collect slightly more device information than I wish they did and are more open to having/using advertising partners than I had expected (from what I know of the service as someone who has never actually used it) but that's like.. pretty tame compared what most of the big platforms have.

Nebula is a shithole, just have a glance at their privacy policy.

https://nebula.tv/privacy

It looks pretty run of the mill to me?

Can access fine (with reduced functionality) on my end with JS disabled - maybe you have something else tripping it up or something?

I used to believe in this (and to a degree still do), but the idea of increasing the attack surface of unreasonable people (who seem to have become increasingly common in the last 10 years) who will do insane things like SWAT you, or doxx your personal details (like home address), or even just follow you around online to harass you has made me have second thoughts about the tradeoffs involved in this approach 🫠

There is a deep irony covering this by writing about it.. on Substack

Thanks for letting us know that your conviction is just performative shoe-pissing.

Okay, in do-or-die wartime I can accept some suppression of the truth.

If you genuinely believe the situation in the US is so bad and analogous to wartime that it's worth spreading and supporting lies for the sake of "winning" then you need to stop wasting your time commenting on nottheonion posts and go figure out how to win. At the very least don't waste time talking to people like me who spend time on truth, set up an automated posting farm and flood the entire internet with anti-Trump propaganda with no attention wasted on what's true or otherwise ethical or moral.

If it's not that bad and you can afford to piss away time here then you can still afford the truth.

Pick a consistent position and take actions that align with whatever you pick, but don't try to have your cake and eat it too, it just comes across as hypocrisy to the rest of us. I hope to either see you around or not see you for a few years -- and nothing in-between.

Accepting whatever makes "the other side" look bad instead of fighting for the truth is the reason we're in this thread talking about that dipshit [Trump]. I'm not sure I'll ever understand why the fuck people in this thread fighting against that so hard

Trump is a convicted felon, promoted beans at the resolute desk, received millions of dollars in inauguration fundraising from big tech, has his own cryptocurrency which he promoted by promising presidential access, and is being called a madman by economists for the least stable US economic policy since Bretton Woods.

Trump is not like "other country leaders". Historically the above would be disqualifying for "other country leaders". You can't cover news about him like "other country leaders". I almost can't believe that you're saying that you can.