JPEG XL

Bruh

I'm using a LG 22mp68vq

It's just 1080p though. But colors look good

If your table depth is enough, buy '24 screen should be great(for 1080)

I'm using viewsonic '24 + rx580 8g

I'm not sure TN, VA, IPS which LCD panel is best for you.

like <@!263309374775230465> use AH-IPS LCD panel, and my LCD panel is VA.

Or you can choose EIZO.

I guess '21 screen means a good screen in 2021 and not a 21" screen. ;)

OLED should be the priority in 2021 if one has the budget.

oh, my bad I just think that('21) is mean 21 Inches 🙂

is that for gaming?

can they dim without PWM now?

and how is the burn-in these days?

I don't have one but I think this was never an issue for typical brightness levels. Though I also never view content with fully white background regardless of display technologie.

I guess it also depends on how often one wants to upgrade

if one is going to upgrade every three years anyway, then it may not be much of a problem

(and by then it might not be OLED anymore anyway)

I didn't get any burn-in (yet?) on my TV

neither did I, but then I only use it as a TV

I don’t spend 10+ hours/day using it

I use it as a computer display, but indeed not that much

Only casual game

Sorry 21' inch

maybe something with G-Sync or Adaptive Sync could make sense

it’s a nice feature

"

How do you feel about the fact that we'll probably have less of new deep fried memes?

(that's a screenshot from June 2018, by the way)

https://youtu.be/QEzhxP-pdos

(I'd like to remake it, but with JPEG XL and Opus codec)

Haha

Should edit the sound to add "XL" then 😅

I'd watch that

> With LAMBDA, Excel has become _Turing-complete_.

oboi.

(https://www.microsoft.com/en-us/research/podcast/advancing-excel-as-a-programming-language-with-andy-gordon-and-simon-peyton-jones/ and https://www.microsoft.com/en-us/research/blog/lambda-the-ultimatae-excel-worksheet-function/)

inb4 `djxl.xlsx`

wasn't excel turing-complete already?

perhaps excel as a whole if you allowed some manual interaction or something like that, not sure

but now the formula language itself is

you can now define functions

and use more advanced data structures than just numbers and strings

ah, right: > In contrast, Felienne Hermans’s lovely blog post about writing a Turing machine in Excel doesn’t, strictly speaking, establish Turing completeness because it uses successive rows for successive states, so the number of steps is limited by the number of rows.

well yes strictly speaking no computer in the world is Turing complete because it has a finite amount of RAM and diskspace so it is a finite state machine

right, but in most programming languages, you don’t have to drag cells with your mouse to extend the computation

Patches is a coding tool in the jxl bitstream that allows you to say "take a rectangle of dimensions WxH at position (x0,y0) from a previously-decoded frame, and put it in the frame I am currently decoding at position (x,y), {replacing | adding to | alpha-blending with | ... } the decoded pixel values"

No, it is also used for still images. The previously-decoded frame can be an invisible reference frame that contains a sprite sheet of letters, for example, that is then used to efficiently encode an image with text.

Patches are waaaay more powerful. They can not only replace (exact matches), but also add (mathematically), alpha-blend, ...

Even now screenshots can use them to improve compression, like this:

https://cdn.discordapp.com/attachments/794206170445119489/834486190576762931/Screenshot_2l.png

(sharp regions are patched, blurry ones are VarDCT-encoded)

Guess Ill try remastering a remastered video with jpeg xl

Just to test other tools

And if i can run at 30 fps

Or Fit into discord

But I cant encode with svt because im still on Windows 7

> Make parts of the image sharper? No, in usual use there should be no visible difference (unless you go really low on quality).

That blurry look of non-patched regions is generated artificially in the decoder, only for demonstrating where the patches are and how do they look.

you can see that the current patch detection heuristics detect quite a few opportunities to use patches, but also still miss quite a few opportunities 🙂

Another example from this server:

https://cdn.discordapp.com/attachments/803645746661425173/811793714112233503/patches.png

What's the image that you want to encode?

No visual difference, only massive size difference

Im was about to ask this

at -d 0 patches does exact matching only, maybe it doesn't find anything at all?

I think patches is Main tool of jpeg xl

There should be minor visual difference (at least on high lossy settings), patches should have less visual artifacts.

But there is no tecnical explanatiom

you can check if you encode with `-v` to see if there's anything in `dictionary`

bits that are in `dictionary` are spent on the patch signaling itself (not the image data that gets referenced, but the references themselves)

lossless now use patches?

Always has been

I didn't know

lossless does use patches by default at speed 5 or slower (higher)

lossy does patches by default at speed 7 or slower

modular work better on text then Vardct (size and quality) and encode 1 time else then 10 is an oblivious size save

cool

but there does seem to be a problem with lossless patches, hm

as in, the encoder does not seem to find any

<@!179701849576833024> did anything change there recently? I'm not getting any patches on the APPLE_BasketBallScreen test image when doing cjxl -q 100, that doesn't seem right...

that doesn't seem right indeed... I'll check

here's you can see what hd resolution screen look like in a computer

saved with paint png of course

is painful

not that i prefer galaxy s4 screen

but galaxy s4 was fine

only it makes a bit of sunskin

so darker skin that makes young people looks older

but now i'm looking to galaxy s4 screen for 5 days

i see how my computer is bad

Is it jxl wallpaper

yes

Fffffff

the jxl wallpaper can be upgraded in resolution

because i have the tree

but the computer you can't do much

other than using more relaxing fonts suited for low resolution screens

like arial etc

or even cabin sometimes can be better

Like wingdings

right

Pwm is not a problem, it's the speed of pwm :p

Is there a tool to display this reference frame? I don't see info about that ^^

What is the magic tricks to separate patch zone to create this image? ^^

Decode with djxl -s 8, so it decodes only the DC and the patches but not the AC.

There isn't afaik, but it shouldn't be hard to make one. Or maybe we already have a benchmark_xl debug image for it? <@179701849576833024> do you know?

I think we have one, but encode-side, and it can be kinda hard to visualize as it might have negative values...

Thank you 🙂

Ah yes, if it's black text on a white background, we actually encode it as negative text on a zero background, so it becomes a solid white background to which the patch gets 'added'

To clarify, this was also demonstrating patches (in response to your suggestion that patches may be infeasible for Japanese text). https://discord.com/channels/794206087879852103/803574970180829194/841168658293325844

Certainly works better in the English comparison with more repetition, but still performs well.

And now I understand why patches don't encode general repetitions well, since they look for contrast.

Do you know what sound elvis presley remixed by JXL ? 😄 https://music.youtube.com/watch?v=Zx1_6F-nCaw&feature=share

we need to collaborate with him :p https://en.wikipedia.org/wiki/Junkie_XL

just learned he composed the scores for the sonic the hedgehog movie and godzilla vs kong

Thanks for the praise in <#794206170445119489>, <@456226577798135808>! More art to come soon… 😉

Should I wait for a fix, or patches on modular lossless doesn't do much?

there is no problem in the sense that lossless isn't lossless or something, but the patch detection doesn't seem to work properly in lossless mode atm and it just doesn't seem to use any patches atm

https://addons.mozilla.org/it/firefox/addon/dark-shatte/

i tried to fix source code as i don't find the firefox colors options

now it doesn't work

You actually did. Turns out we had an encoder bug since end of November that caused the encoder not to detect any patches if the colorspace is RGB (i.e. in lossless mode; in lossy we use XYB by default).

It will be fixed in the next sync, <@179701849576833024> found and fixed the encoder bug.

So <@111445179587624960> you may want to re-run some lossless benchmarks after the next sync happens, for some kinds of images it could make a significant difference!

Remarkable that the bug went unnoticed for nearly half a year, but the encoder was doing nothing wrong, it just didn't detect patches in lossless mode and didn't end up using them.

Ok, did the patches not work even when enabled in the options? As far as I remember in my tests sometimes the results were different (but maybe it was on quite old builds)

Btw, for the new re-run, it would also be nice to have an option for invisible alpha

I thought the option was there

at least for lossy, not sure if <@!794205442175402004> decided to expose it 😄

Yes, I mean for lossless as in FLIF/WebP(2)

In lossy something is always done to the invisible pixels, not making them black but smearing visible pixels into a blurry mess in the invisible region, which is easy to compress (blurry messes have low entropy) and is better to avoid artifacts from invisible pixels bleeding into the visible ones.

In lossless currently it just doesn't touch invisible pixels at all.

And premultiplied alpha is also not yet exposed as a choice, we just do it for EXR input and not for any other input.

No, they didn't work. As far as I can tell, patches still worked in version 0.1, but were broken (in the lossless encoder) in version 0.2 and later.

In version 0.4 they will be fixed again 😅

I mean, they might in theory have found something in 0.2-0.3, but for sure they didn't do as good as a search as they could have

There are so many incredible paper on video edition: https://www.youtube.com/watch?v=P1IcaBn3ej0

Better lighting = more realistic game got it <:kekw:808717074305122316>

here it's better than ray tracing :p

Actually, I wouldn't say so.

Except for the encoding time, because I am using a PC at that time and it will be inaccurate

Wow they played GTA V on xbox

their footage is all very green

it doesn’t look like an improvement to me

look like a real bad smatphone video

but look more "realistic" in this manner

at the very least, it might look better if the data set was properly white-balanced

not sure why they left that strong green cast

it’s all I can see

Yes, it has a strong green cast, and also looks undersaturated (the synthetic source looks a bit oversaturated though)

Yes but with with balance correction it look good. I don't know why they keep horrible dash cam white balance.

It shows the need to prune your training data carefully

If you want a specific result

the road also looks unnaturally smooth

But that's how it looks in reality though, the GTA original is too sharp, like if someone dragged Photoshop's "Clarity" slider up to 11

it doesn’t look _that_ smooth irl

I think the problem is that the video is really compressed and low resolution.

The problem with the video is that most of the things done in the video can be done using competent Reshade tweaking <:Thonk:805904896879493180>

Because you've probably only watched poorly compressed YT video. See static images here, the quality is so much better: https://intel-isl.github.io/PhotorealismEnhancement/

better indeed, thanks

The colors looks like nfs most wanted

How to convert GTA to nfs most wanted :p

It is very much off-topic

The Dutch library foundation reviewed my science fiction book

Haha no, some libraries are purchasing my weird book now, which is kinda cool

https://sneyers.info/boek/ is the book

It's an epic optimistic-dystopian feminist pornographic hard-SF adventure, written in Dutch.

Haha well it's just words, obviously

Writing for both humans and computers. That's similar. 🙂

Not quite, haha. Computers are easier to write for, imo. Human languages are much more complicated.

there have been attempts to make human language more computer-y, for example with lojban

.i ma tavla la lojban

I doubt such attempts will have any significant success. Human languages tend to evolve naturally, and normative linguistics is usually seen as something outdated - basically they've mostly given up on it, and everything is becoming descriptive (describing the weird stuff humans do) instead of proscriptive (demanding that humans do less weird stuff).

But AFAIK for example Esperanto (which is at least 7× easier to learn than English) was on the rise and then some nations blocked it because they would lose money on spreading their languages into the world. So it again leads to money and power, like many other big issues in our society…

I mean… not the whole nations but their establishmens.

7x... that's a remarkably specific number

I actually had an “argument” about this in a YouTube comment section recently, where someone was arguing very pedantically that > An English accent in English, a French accent in French, a Chinese accent in Chinese, and a Korean accent in Korean is by definition the same as saying someone is speaking properly or accentlessly.

had to explain that according to the definition of “accent” that linguists use, everybody has one

and that there isn’t just one French accent, for that matter

(neither is there even just one Chinese language… but I didn’t feel like going there)

There was a lot of research on comparing constructed and natural languages. AFAICR some concluded 7×, some even 10×.

Italian also definitely has maaaany different accents (and dialects, too, but that's another story)

yep, and AFAIK this counts as dialects in linguist lingo

> The most significant variations or differences within languages occur at the level of the lexicon (vocabulary), phonology (pronunciation), grammar (morphology and syntax). and usage. Moreover, they are not just qualitative, in the sense that dialect A uses one feature and dialect B another, but they may also be quantitative, in the sense that dialect A uses one feature more often than dialect B does. (This is particularly true of phonological and grammatical features which have social or stylistic significance.) Finally, variation may be regional, social or stylistic in its origins, and the methods that linguists have used to study each type differ slightly.

FYI this is from http://www.stanford.edu/~rickford/ling73/reading1.doc (yep, sorry, it’s a .doc), which, as a layman, I found a valuable summary of how linguists approach the subject

(I mean, the title is “How Linguists Approach the Study of Language and Dialect”)

to me, an interesting paradox is that the study of people’s attitude towards language is _also_ descriptive itself

so if people’s attitude is prescriptive, as a linguist, you wouldn’t prescribe that it should be descriptive instead

and in a sense, Perl was an attempt at doing the opposite (but not in the English-like way of, say, AppleScript or Adobe Lingo)

Larry Wall is a linguist, after all

kind of relevant: https://twitter.com/thejazzemu/status/1384192060631093250

COBOL was kind of an attempt to make programming languages more Englishy

It sucked of course

Python is a lot more "Englishy"

I like C++, mostly like rust, and have a love-hate relationship with Python 😛 (and with stuff like bash, but that's less of a "programming language" in my opinion)

I have been doing too much logic programming in the past (Prolog and stuff), to me all imperative programming languages look kind of the same, all just different flavors of syntactic sugar and stdlibs for machine code.

I haven’t actually used Haskell for a lot of “useful” things but I have kind of a soft spot for it

I like it in principle

raku (formerly perl6) can be fun too

yup

My arm hurts a bit, but nothing unbearable :slight_smile:

Congrats!

looking forward to mine 😄

(it’s on Saturday)

(as in, in 38 hours)

Mine is in -688 hours, and another one in 19 hours!

Mine didn't hurt at all which was kinda surprising…

Sorry if my (now deleted) post seemed offensive. It wasn't meant as such.

FBI subpoenaed Sci-Hub's founder's Apple account. If you've got some disk space, seed, because Sci-Hub may get taken down soon...

https://www.reddit.com/r/DataHoarder/comments/nc27fv/rescue_mission_for_scihub_and_open_science_we_are/

This is garbage: most of that research is funded publicly, but these publishing corporations get to double dip for profit. It's abhorrent.

I prefer not to comment on that xD

my stance of paywalls for research papers is about the same as paywalls for standards

https://www.youtube.com/watch?v=Yq3P-LhlcQo you might recognize a name or two that appear here 😄

Amazing stuff! How did you do the counting, <@!795684063032901642> and <@!179701849576833024> ? Brute-force all the way through an enumeration of all the trees? How much cpu time did that take? 😅

a *bit* better than brute force 😛 the trees are not quite the problem, there's not that many

https://github.com/google-research/google-research/tree/master/cube_unfoldings

up to 9d it can run on a normal-if-powerful PC in a day or so (10d needed a cluster)

(read: something like my desktop, threadripper 3970x with ... enough... ram :P)

is your desktop the normal-if-powerful PC or the cluster here? 🙂

hahah no the cluster has significantly more oomph than my desktop

https://oeis.org/A091159 still needs to be updated with two more terms 🙂

so the 4D ones have 3D unfoldings that all tile 3D space?

do the (n+1)D ones have nD unfoldings that all tile nD space?

https://tenor.com/view/jason-mantzoukas-betting-gamble-place-your-bets-the-house-gif-8225040

I don't know

but it would take quite a bit of computation...

that or some kind of ingenious proof by induction, I guess

Is there any connection between the tilings of 3D unfoldings and the tilings of 2D unfoldings?

also: the video showed how the Dali unfolding itself has a 2D unfolding that tiles. Is that true for the other 260 3D unfoldings too?

or not by induction, yes (I somehow don't think induction would help much here)

I don't *think* so

no idea... <@!795684063032901642>

no idea, the question also seems a bit far-fetched

Is there a tiling pattern that uses all 261 unfoldings?

(I can come up with more far-fetched questions if you like)

i don't know what this conversation is about but i saw somebody writing something about programming language preferences ect and the right answer is C

(just saying)

C is p good but not always the right tool for the job (proglang pref discussions are silly)

https://www.youtube.com/watch?v=tas0O586t80

ugh how do you track down licenses for things that kind of look like they're probably CC0 or CC BY-SA or something, but the author/license seems to be very hard to find?

https://web.archive.org/web/20111221090506mp_/http://s826.photobucket.com/albums/zz190/apng/?action=view&current=APNG-Icos4D.png

google reverse image search may help?

id just claim fair use tbh lol

I suspect that would not fly for Official Test Corpus though

Just over throw the berne convention and institute a new era of free media! <:galaxybrain:821831336372338729> ☭

For works by anonymous authors, the copyright is 95 years lol

Even if it's abandoned, it can't be legally copied outside of fair use for 95 years

https://xkcd.com/2464/

> simple and here I am with my mere 3D brain

Complete guess: some extension tag that's expected to be appended to by every processing step?

I would expect that there might be a way to tile that uses each tile at least once, but I am not so sure that there will be one where all the tiles are used equally often. So I don't really expect that you can arrange the 261 pieces in order to obtain a "prototile", which tiles the plane. In any case I think a somewhat practical question would be: What is the smallest box, such that you can fit all the 261 pieces in it. A perfect fit (and also a postive answer to your question) would be a box of size `29*12*6`, for example. But can we even pack the pieces in a slighly larger box, e.g. `7*12*25` or `13*13*13` or something? This would be useful in order to 3d-print all of the pieces without using too much space. When we want to print both versions for the chrial ones, it is actually 455 pieces.

Microsoft Word feels very demented

I click "Accept all and stop tracking" in a document with 17k track changes

it starts accepting a few dozen changes, slowly, and then after a few minutes it forgets what it was doing and just stops

clicked the button a few dozen times now, am down to 11.6k track changes

I would use latex for this if I had a choice

but I don't

ISO requires f'ing MICROSOFT WORD and they don't accept anything else

markdown >> latex

markdown? seems a bit weak to write a spec in

maybe not though

can it do cross-references?

not with the right extension

or save in docx

no, you get a docx from the ISO editor, and you're not allowed to convert it to something else and back because then the magic xml stuff they have hidden inside the docx would disappear

or something

onlyoffice?

wps office use also some word magic

(i'm just joking)

docx is open

anyway, it annoys me even more that ISO puts everything behind a paywall

I'm a discord/code noob, someone can explain me all this github/discord anime coulture?

Damn I'm fabian

With anime your code runs 10x faster.

I must try with jojo

OOXML is an ISO standard....

I’ve used reStructuredText for a while, it was rather nice

one nice thing about those is that it’s quite easy to generate accessible output in, say, HTML or EPUB

PDFs generated by LaTeX look very neat but I have heard that they are typically not that accessible, are they?

I have no idea, I'm not a member of ISO

or otherwise, we could make it more bureaucratic to please ISO, and write in DocBook

There was significant pushback because OASIS (.odt) had already been standardized

How do you write math equations on markdown. On latex, it was very easy.

it’s XML, which means it must be good

as everyone knows

you can write them in LaTeX, actually

if you convert from markdown to LaTeX, you can keep them as is, and if outputting to HTML, you can use e.g. https://www.mathjax.org/

In latex syntax

You can test on hackmd.io

So how's the 5g reception now?

Unchanged :(

I got the J&J vaccine and I only get 4g reception <:FeelsSadMan:808221433243107338>

So need latex. Latex > markdown :p

is a better latex

at display time if outputting to HTML, MathJax can convert the math to MathML (or render it itself), and there won’t be any need to parse all the non-math LaTeX and map the rendering of _that_ to something semantic

having taken notes in reStructuredText (with LaTeX syntax for formulas) for 2-3 years in university, I found it to be a more straightforward experience than using LaTeX for the whole document

Wait nobody is using JPEG XL but WebP2 and TIFF are viable?

People are weird

But jxl has probably the best lossless compression ratio

Also

Wait

>dropped jxl because nobody is using it >webp2 is viable

What

Also tiff is funny because if you're using tiff you may as well use PNG lmao

Is that comparable to PNG?

Also wouldn't zstd in tiff be extremely incompatible with practically everything

I say for lossless if you want compatibility you choose PNG, if you have time you zopflinate it, if you want better ratios for less compatibility you use webp lossless, and if you want best compression and least compatibility as of now you use jxl

Webp lossless is a good compromise I think

At least for now

Once jxl becomes enabled in a newer chromium it becomes viable for most of the web and usage in general

For data science, being able to deal with more than 8-bit is nice, I'd think

Yes maybe that's why tiff is used

But PNG supports up to 16 bit right?

Yes

And WebP is limited to 8-bit, WebP2 to 10-bit...

If the format is for internal use then tiff with zstd is suited

I think

But the argument of "nobody is using it" is moot

Because adoption for an internal format matters almost none

A more legitimate issue would be speed concerns and CPU concerns

I could understand not using it for that reason

But that's not the reason they give

Shouldn't we invite GerHobbelt to this server?

If you can get him to come <:Hypers:808826266060193874>

The more (developers) the merrier

That would be a legitimate concern

But adoption isn't the issue

Jxl isn't perfect for all scenarios especially in lossless scenarios with limited resources

Not really worth switching yet, not enough enabled-jxl browsers out there yet...

Lossless with limited resources: it would be possible to make an -s 3 style encoder that uses constant memory (assuming you can feed it the input image in group-sized tiles)

As long as you don't use Squeeze or delta palette with WP, and don't try to make a global MA tree, groups are independent and can be encoded and decoded separately.

to be fair, TIFF supports floating point too

AFAIK PNG and WebP don’t

(but JPEG XL does)

Tiff supports 32bit . PNG doesn't.

That was also asked and veluca said it is possible

But the priority is adding s2

Or faster decoding

He mentions more faster decoding

when there's the option at all, yes 😄

ANOTHER TEXT IN R/AV1 SERVER

Frikandelbroodje met extra curry

<:Frikandelbroodje:827322586832175184>

Yes. It passes for me.

Well, I'm on Samsung Galaxy Note 9, so I also have physical Samsung Knox "e-fuse" and the only way to fix it is to replace the whole motherboard, so no root and different ROM for me, I guess...

I bought it just for the S Pen. Believe me, it's ingenious and if not that, I'd move to another phone or maybe even another company.

on the positive side, they have now committed to providing security updates for at least 4 years for new products: https://www.samsungmobilepress.com/pressreleases/samsung-takes-galaxy-security-to-the-next-level-by-extending-updates

I think Google is trying to separate image codecs & video codecs from OS on future Android releases. So they can be updated from playstore.

They are doing this, so they can patch vulnerabilities.

❤️

no is because i've deleted messages and you use only hims

Android 12 will be released on September or October. Is it possible to ask them to include jxl in Android 12 ?

nope

but it can be in apps

<@416586441058025472> continuing from <#794206170445119489> there is a Instagram mod called instander, which allows you to bypass Instagram resizing / re-encoding so you can upload high quality photos.

Instagram supports dci p3. If your phone camera app support encoding in dci p3, You could use that. I see better colors when using p3 .

dci p3 or display p3? I think display p3 is more common (it's p3 primaries with the usual sRGB transfer function)

(on phones etc I mean)

I don't know thr exact name. It shows wide color or something in camera app settings.

Both exist, but dci p3 is used mostly in the movie industry while display p3 is what Apple started doing and I think also other phones/screens

Then it should be display p3. I will check it again.

if there's one thing I learned working on jpeg xl, it's that color is *a lot* more complicated than one would usually think...

Mostly needlessly complicated but yes

It says dci-p3. Not display p3

that's... peculiar

ah, apparently DCI-P3 *can* be the name of the combination of primaries and whitepoints...

Also phone display specs say its dci-p3 compatible

anyway the transfer curve is just a 2.6 gamma, so should be easy enough

DCI-P3’s white point is also a weird greenish tone

at ~6300K

(but greener than “D63”)

the curve shows daylight chromaticities, the cross is DCI-P3’s white point

<@!416586441058025472> Did fabian patent jpegxl? 🤔

I think Fabian meant that JXL is patented and that's bad(?)

or patented things cannot be defended

The patent system is sad, but the way it works is that the best way to make a royalty-free codec is actually to have defensive patents in it (granted royalty-free with a defensive clause).

https://music.youtube.com/watch?v=lqda0KL7Tgg&feature=share

All theese ball have the same color :D brain do SO MANY visual simplifications :D

pretty fancy as far as optical illusions go...

This one is pretty mind-boggling. It even works with my eyes super close to the screen.

This image is a good argument in favor of chroma subsampling

(I can also give some good counterarguments though)

no they are are green blue and orange

Cool indeed. The guy just shouldn't use lossy compression for that kind of image.

I always say that all kids in elementary schools should be taught the difference and usage of lossy and lossless compression. 1 hour would be enough.

ah are you saying in reality those circles have same colour?

yeah they're the same color, i find it easier to see if i zoom in a lot

Yeah they are in same color.

No, Fabian is right!

I checked.

e.g. one ball has hue 27.234589

another is 27.823429

That's the dark side of lossy… 😉

if you recompress that image often enough, the color of the stripes will actually bleed into the balls

It's david novick files: https://twitter.com/NovickProf

PNG file 😮

He create so many files 😄 https://www.dropbox.com/sh/oi9wdd0uc5uc700/AAAXFo4uWPG6ZEygQ5GY50tma?dl=0.

it's interesting to see which color pattern is the best to mix color by the brain 😄

He change color when he zoom to keep the same visual color perception 😄 https://www.dropbox.com/sh/oi9wdd0uc5uc700/AAAXFo4uWPG6ZEygQ5GY50tma?dl=0.&preview=Dynamic.mp4

in motion brain is better 😄 https://www.dropbox.com/sh/oi9wdd0uc5uc700/AAAXFo4uWPG6ZEygQ5GY50tma?dl=0.&preview=Curtain+1.mov

https://youtu.be/Ot8YhBWUfGs

It speeds up when carried.

if you nearest neighbor these a lot they actually look like the illusion<@!231086792315633664>

especially a downscale and an upscale with AI will make them look like the illusion

<:Hypers:808826266060193874>

using a bunch of downscales and AI upscales I got this result

looks exactly how it seems at first

Would be funny input for a generarion loss test

By funny you mean excruciating?

I just found JPEG XL because of the new german copyright law.

So you are allowed to share images without problems if they are <125kb

what?

that's an... interesting... threshold

It's the "meme" paragraph.

Uh

So much for copyrighting jxl-art

So better compression means fewer copyrighted images?

That is an interesting new motivation to do compression research, haha

Oh not exactly like that

If you post a image on your 100k visits news page you can still get in trouble, but you should ask not me for the details.

You might be able to classify jxl-art as vector graphics.

that'd be some interesting classification criterion then xD

But that might imply JPEGs are vector graphics.

So if you want to give some lawyers a headache...

<:WhatThe:806133036059197491>

If I upload images to Cloudinary, which countries' law do I have to follow? The one where Cloudinary has its headquarters? The one where Cloudinary has its servers storing my uploads? The one I'm uploading from? The one my IP address is from? The one where "cloudinary.com" was registered? All of them?

Good question

It might be the country you are a citizen of

At least for some aspects of it

For some aspects it might be USA

Probably best not to do things that are illegal in USA or illegal in your own country

IANAL, but I believe generally businesses need to comply with both laws of jurisdictions they are registered in, plus the laws of jurisdictions of customers they do business with.

<@456226577798135808> That was not a good idea...

Goodbye <@456226577798135808> 👋

You will be banned from this server <:SadOrange:806131742636507177>

It was nice to see you @everyone <:kekw:808717074305122316>

wait anyone can ping <:monkaMega:809252622900789269>

<:monkaMega:809252622900789269>

@everyo— _\*banned\*_

Anyone can, but ziemek.z is the only one brave enough to do it multiple times.

`@anyone` who wants to play ping roulette?

I have been pinged again

Only two times, to be exact, but it still technically counts as multiple, so...

quite bad

does voteban work even?

!votekick

!voteban

nope

lol

https://tenor.com/view/lotr-lord-of-the-rings-theoden-king-of-rohan-you-have-no-power-here-gif-4952489

<@!179701849576833024> I think it's time to remove pinging permissions in regards to @ everyone and @ here.

yeah.......

@anyone

mods are asleep :^)

it's a wonder sophia or fabian hasn't discovered this power

I've only pinged everyone 2 times. Now imagine Fabian doing that

Ultimately, this will be a productive troll.

Actually, I think it's time to give me mod role <:ReeCat:806087208678588437>

Fabian is here?

He's everywhere <:PepeOK:805388754545934396>

He's all encompassing and all viewing.

i'm going to do it

<:Stonks:806137886726553651>

<@!416586441058025472>

I hope he doesn't see into my restroom window.

NOOOOOOO

I wonder if this is still a thing.

https://youtu.be/WGDU5xifwRQ

gods that video is great

Aww and here I was thinking people were messaging me!

(to be clear, I wasn't pinging @ here, as you can see in the message)

<@!336697008356327444>

hi

also it doesn't work if you backslash

What doesn't work?

That was surprisingly spammy 🤣

the @ everyone warning

Mm. I was pinging a user though.

So @ here is supposed to only ping people in that channel who are online, while @ everyone pings the whole server

Only devs and core devs can do that now

And probably shouldn't, except if there's some really breaking news like "safari is adding jxl support" or "major security bug found, malicious jxl file can install virus on chrome"

"jxl turns out to be turing complete; cjxl accidentally proves P=NP"

I am somewhat tempted to do one global ping just because xD but maybe just pinging <@456226577798135808> a few times is a good revenge anyway...