JPEG XL

And would be able to correct exactly one bit error

so rotational velocidensity affects jpeg xl

my view is that bit rot is something that needs to be addressed at a different layer than the codec itself

it's something that should be implemented in storage media and transfer methods anyway, and has to be adjusted to the amount and pattern of errors that can be expected

yeah i struggle to see how you end up with a corrupted file in the first place

There is value is having a check for corruption, but correction should be done at a different layer.

You would need very different capabilities in a datacenter environment vs a space probe, for example.

and if JXL is used as an archival master vs. for transferring a lossy version

I mean, not really, depends on how much you're willing to pay for it in computational cost... I believe you can always use O(# bits to fix) to correct the error, at the cost of potentially a lot of compute

If the compute is equivalent to mining 50 bitcoins, it's not really useful, I guess

Even error detection codes can be used for correction: just try flipping bits until it decodes again.

I mean, it's "only" quadratic xD

but no, that only works if you can detect *two* errors

I love it.

Still £30 to find out how big screws should be though: https://committee.iso.org/standard/3707.html

what?

He wants ban from a subreddit I think

Ban does it mean i wont see the content and images

(i think you were talking about reddit) you can still see posts, just not post something yourself

I don't understand why new Qualcomm adding 18bit support for camera. Are they stupid or what

18bit?

For HDR it might make sense, I suppose

Spider Mario said 14 bit is good enough.

Still no av01 hw decoder <:Hsss:806131225278152756>

I think for displays there is not much value in more than 12 bits, but for cameras there can be quite a lot, as it can allow a greater dynamic range if something really bright (like the sun) is in frame.

https://discord.com/channels/794206087879852103/794206170445119489/883746018498330644

For a single exposure, 14 bit is plenty, but I can imagine modern cameras with multiple sensors doing multiple exposures and actually getting close to 18 bits worth of precision/range out of it.

https://twitter.com/LymeScience/status/1467977563423318025 oddly specific

legōrum

wait no i forgot how latin works

legae?

legō could be the dative or ablative singular of "legus/legī", in which case the corresponding (dative and ablative) plurals would be "legīs"

the plural is lego <:YEP:808828808127971399>

legos

Legoes

One lego, two legoi?

Maybe it's uncountable, like water

two waters

Five liters of lego

legolas? xD

https://c.tenor.com/mmFKcPzBAUUAAAAM/legolas-lord-of-the-rings.gif

See also, *foot pain inducers*.

What is it with websites and having images with insane resolutions and displaying them in small res? <:KekDog:884736660376535040>

Like this: https://www.shimmercat.com/

5000x3337... but the displayed image is small XD

Rendered res: 540x360

What's really bad is that the product is meant to optimize images. Ouch. They either have a really bad marketing team that didn't optimize images or they use it as a recruiting tool for people that report it? No idea.

on the other hand, it’s a heavily compressed AVIF (371 551 bytes, -> 0.18 bpp), and the rendered resolution depends on the display density (on the 8K monitor I am trying, it is rendered at 1620×1080)

although the JPEG version _is_ 1 572 494 bytes

(0.75 bpp)

But that's 8k which is an extremely small market and even then it's way bigger than the rendered res :p

see also: UK Type G Plug / BS 1363

The counting of the Kshama Sawant recall election is a real nailbiter

Obviously I am rooting for Kshama

This quality <:KekDog:884736660376535040>

Looks nice https://youtu.be/kBsNvGyfM8Y

Ah yes, rain and reflections

At least some realistic amount is nice unlike some other mods where literally everything is just wet and shiny

this would look so good in HDR

but then the display might overheat too 😄

https://youtu.be/BpnnD_0IlbE

Have anyone ever tried this gpt-3?

This gpt-3 thingy lol

>not compressing your audio by converting PCM to bitmap first

This is what they use to write those garbage bot written articles.

>not compressing your videos with opus by converting frames to PCM first

"HEVC is more efficient than H.264, but it is not as efficient as Opus." the most cursed sentence I've seen this week

https://jacobinmag.com/2021/12/seattle-kshama-sawant-reelection-socialism-city-council-capitalist-strategy/?utm_source=dlvr.it&utm_medium=twitter

https://i.redd.it/zrv64iiagw481.jpg

There isn't any curved line, right? Strong illusion...

Cool

https://www.rt.com/news/542904-new-zealand-covid-vaccination/

RT News <:monkaMega:809252622900789269>

I found it on YouTube

i'd just use a scraper like mangakakalot

i have .com bookmarked

mangadex just sucks

for my yuri i just get straight from dynasty and i have my own mirror for LL

their authority is a problem, their ads are replaced with "sponsorships", they're not transparent about their funding, their website is down half the time, the SPA sucks, the server sucks

scanlation is still piracy yet they dont allow english licensed uploads because it's piracy 🤔

yeah

the website is ass

the funniest part imo is the search "box" that is actually a button that opens a dialogue

is hitomi.la only hentai?

most of the doujins are sfw

and some groups only upload to dynasty/upload to dynasty first

so maybe

i should have 1 semester left

idk about final year projects

i think my final year has been even less busy

that's not because piracy, but because there are parties selling stuff who care. in absence of these you don't harm anyone because no licensed party captures the market (english readers)

because there's no harm, piracy is allowed ok

because publishing on behalf of another person is ok as long as they don't understand it

we've seen many examples of authors tweeting against scanlations. there isn't no harm

it's more about nature of market capture & who you damage & whether the one you damage have power to sue

power not only as legal power but also about care to sniff out who uses what and effort sending dmca and legal threats

in that sense, there's no damage when pirating anime in the west either (except for Netflix)

but the people who argue for the manga don't think so

the piracy argument is *I wouldn't have bought it anyway*. that's the market capture

tbh if author have no plans to release in english then i dont see why they would care about translated stuff. for raws, i can see the point, as intended market is the same as one you are already supplying

Md should just ignore DMCA

yes, there is some value lost if stuff gets released and you don't control that market, but there is also quite a lot of effort put into translation work itself

not allowing some series to be uploaded also hurts the ux

ignoring DMCA opens you to real legal threats, if submitter has a case

just move to a country that can ignore dmca

like canada

and in cases of licensed works, they usually do

and its not sole translators, it's whole companies so they have framework for real legal stuff

i mean you can ignore DMCA but what comes next aint nice

canada doesn't ignore copyrights

DMCA is actually the nice option when it comes to this

it's hard to believe that manga distributors have more sophisticated legal than anime

As for MD, licensed stuff is only half of it

id say its only problem because they wanna operate very accessible site and do it in legally okay way

they also decided to give scanlators rights which means groups take off their stuff when they feel like it

there are hella lot of examples of sites which either operate in less legally okay way

or being less accessible so they are less known

i used to source my stuff off manga.madokami.al but i lost credentials to that :<

I hate doki

it's often forgotten that md was anidex is sdbx is doki

Trying to be an accessible site is reasonable

I don't think it ended up being towards anyone's benefit

meanwhile i found something <https://ftp.pptenshi.live/README.txt>. stuff i care about seems to not have latest chapters though and no clue how long this will be up

I remember that

uploaded a ton of garbage when it was first created

thought it died because of the garbage

https://news.cnrs.fr/articles/seeing-in-the-dark

o yeah infrared and even bright red light have positive effects for seeing stuff

seems infrared even more so due to oxygen stuff as described, but even red light can help by warming up relevant cells and thus improving blood circulation there

so basically looking at fire is a good thing for eyes

rude. i just remembered some stuff ive read about red light elsewhere. https://www.usnews.com/news/health-news/articles/2021-11-29/red-light-in-morning-may-protect-fading-eyesight-study

or, well, more direct https://www.nature.com/articles/s41598-021-02311-1

it's pretty clear that more oxygen to relevant cells = eyes working better, and that fire is full of both red light and infrared so i don't think i said anything wrong here either

Call of Juarez Gun Slinger is free until December 14th: https://store.steampowered.com/agecheck/app/204450/

is that (final) use case even relevant by now? I always used scp for years.

iirc ftp/ftps' ways of doing things (eg. single tcp connection per file transfer) are faster due to the way ssh multiplexes stuff and handles buffers. there are things like <https://www.psc.edu/hpn-ssh-home/> but they aren't upstreamed afaik

also anonymous sftp setup isn't very clear compared to what some popular ftp packages provide (but then some of these packages provide sftp protocol too so this point may be moot)

<@416586441058025472> yo. <@219525188818042881> is wondering when you think opus 1.4 is coming out

Lol

Wouldn't Blue know more about that?

Fabian is the audio encoding king tho

https://www.youtube.com/playlist?list=PLUF3AINVFJxru_zAMizWwb_1mVTJlcshi

https://www.youtube.com/watch?v=WU0gvPcc3jQ

https://github.com/veluca93/libjxl/network/members

Random question, but does anyone know how big the difference between guetzli and mozjpeg is on average?

Tried it on 4 images. With 2 mozjpeg completely destroyed guetzli (guetzli smoothened it too much and mozjpeg didn't produce noticeable artifacting) and with 2 other images guetzli won, because it looked visually more appealing

generally I think guetzli is better for q90+ and mozjpeg better for q<90, roughly speaking

Hm, guess I'll use that as target then if I want to get the best efficiency since encoding time usually doesn't matter at all. Hoped and expected guetzli to be better at q84 and higher

I only tried q84, so that might explain the wildly varying results. Hoped Guetzli to be 1% - 5% better (with optimization since optimizing Guetzli encoded images can greatly reduce the size) in the vast majority of cases... especially since it takes hundreds of times longer to encode

Once optimizing it made it 8% smaller <:KekDog:884736660376535040>

I've had one that provided a sandboxed SSH

I don't remember, it was *years* ago

now I pay $5/mo for a VPS and get full SSH access

same, 3.22€/month at hetzner.de

(+ separate domain names)

domain names are super cheap

I don't use Hetzner since I'm in north america. I use Genesis Hosting which is in Chicago

depends on what one wants :p but .com, .org and a few others, generally yes

https://www.da.vidbuchanan.co.uk/widgets/pngdiff/

Found this on HN

https://discord.com/channels/794206087879852103/805176455658733570/920895596703907850

https://www.youtube.com/watch?v=zX21-Lqsj3s

Val Thorens https://pbs.twimg.com/media/FGzShBuVUAM0x8I?format=jpg > https://mobile.twitter.com/GenshinImpact/status/1471782706270769153

https://github.com/jiaxi-jiang/FBCNN

I think this will work on h264 too.

gonna post this to iew

we already have dpir

https://www.ikea.cn/cn/zh/p/blahaj-bu-luo-ai-mao-rong-wan-ju-sha-yu-10373589/ https://www.ikea.com.hk/zh/products/play/toys-for-small-children/blahaj-art-10373589 https://www.ikea.com.tw/zh/products/play/toys-for-small-children/blahaj-art-10373589 https://www.ikea.com/jp/ja/p/blahaj-soft-toy-shark-10373589/ https://www.ikea.com/kr/ko/p/blahaj-soft-toy-shark-10373589/ https://www.ikea.com/us/en/p/blahaj-soft-toy-shark-90373590/ https://www.ikea.com/ca/en/p/blahaj-soft-toy-shark-90373590/ https://www.ikea.com/mx/en/p/blahaj-soft-toy-shark-90373590/ https://www.ikea.com/dk/da/p/blahaj-tojdyr-haj-30373588/ https://www.ikea.com/de/de/p/blahaj-stoffspielzeug-hai-30373588/ https://www.ikea.com/se/sv/p/blahaj-mjukleksak-haj-30373588/ https://www.ikea.com/no/no/p/blahaj-toyleke-hai-30373588/ https://www.ikea.com/it/it/p/blahaj-giocattolo-morbido-squalo-30373588/ https://www.ikea.com/ua/uk/p/blahaj-blohey-igrashka-m-yaka-akula-30373588/ https://www.ikea.com/cz/cs/p/blahaj-plysova-hracka-zralok-30373588/ https://www.ikea.is/products/577037 https://www.ikea.com/my/en/p/blahaj-soft-toy-shark-10373589/ https://www.ikea.com/sg/en/p/blahaj-soft-toy-shark-10373589/ https://www.ikea.co.id/in/produk/bermain-dan-belajar/mainan-anak-kecil/blahaj-art-10373589 https://www.ikea.com/ph/en/p/blahaj-soft-toy-shark-10373589/ https://www.ikea.com/au/en/p/blahaj-soft-toy-shark-10373589/

Shopping list?

shopping list for cool people <:YEP:808828808127971399>

Your favourite toy?

nope, it's everyone's favourite toy 🙏

I think this is kind of important to read: https://www.theguardian.com/world/2021/dec/18/scientists-watch-giant-doomsday-glacier-in-antarctica-with-concern

Seeing our politicians argue back and forth over the issue for the last 20 or so years, and somehow retreating from their stances from 20 years ago, I'm doubtful they'll do anything

As it stands their official policy is "We're sure to figure out a way to fix it in the future so why change things now?"

The problem with politicians is that generally, they can't look further than (at most) ~4 years into the future. Anything after that is after the next elections so they don't care.

(similarly, most companies don't look much further than the profit numbers for the next quarter)

Some do: camera raw and PNG are being used. But on the web, there's also big pressure to 'optimize' and images are a big fraction of median page weight...

Lol "To be specific, the results from BBC R&D found that VVC offered up to 27% bandwidth efficiency over HEVC at HD quality and 35% better at UHD quality. For example, a 90-minutes 4K video encoded HEVC may take 10GB, but it now needs only 5GB in H.266 codec."

"35%" "half" <:SovietThinking:821039067843133462>

I wonder how many frames per minute it encodes

or maybe minutes per frame

https://github.com/Arisotura/melonDS/issues/1296 “interesting” thread

Wonder if he his knowledge stops at turning on the PC since stuff he says doesn't make any sense at all XD

I enjoyed this bit: > and if cmake is all that powerful why it can't generate a .bat file?

Don't know anything about compiling, so no idea why that's funny, but since he said it it must be wrong

Like this: "well i guess my question gave my intentions clearly that i dont plan to downgrade my OS to 64bit"

"will never use 64bit crap"

"if MS had used NT the way it was designed they would never had needed to force people to use win10 to have newer features"

What on earth does that mean

that's interesting, accepting 32-bit but not 64-bit

why not stick with 8-bit or 16-bit?

I DEMAND LIBJXL TO WORK ON COMMODORE 64

yes, they seemed to express love for DOS and Win16

> can't believe you guys don't know how much of a scam 64bit is, I'm running Windows 11 in 16-bit mode and it only uses 2mb ram <:yep:721359241113370664>

> oh lord you guys use qt5 i think i dont want to try this anymore...

Minecraft texture pack creators be like: let's make them PNG-32 What's needed for exactly the same visuals for many: PNG-4

Quality optimizations <:KekDog:884736660376535040> (with stripping the metadata, reducing the color depth and optimizing them, they actually became about 50% smaller from a test on 1 pack XD)

oh is this why people say that 2001 was the first year of the 21st century?

is what why people say that?

for 2021 to be the 21st year of the 21st century, 2001 has to be the 1st year of the 21st century

yes

it is

2000 is the 0th

https://en.wikipedia.org/wiki/21st_century > The 21st (twenty-first) century is the current century in the Anno Domini era or Common Era, under the Gregorian calendar. It began on January 1, 2001 (MMI) and will end on December 31, 2100 (MMC).[1]

i'll allow it

but does it start at 00:00 or 00:01 clock?

It starts exactly when 11:59pm becomes 12:00am

Random huge AI enhanced jxl image <:KekDog:884736660376535040>

https://twitter.com/ResNeXtGuesser/status/1473141189922623495

<@!111445179587624960> Which version was fjxl you shared?

Latest build

what's the bug? xD

new.png doesn't look quite as lossless as org.png.

Original PNG:

FJXL (as PNG):

Hmm, yep

tl;dr fast lossless bite off a part of my pixel cookie

Interesting, I will look into this when I'm no longer on a train :P

I fixed your cookie 😉 (even if I'm still on a train... :P)

A next-gen train development

> and do what live forever on a older release? nah i was hoping too much with this project... but i tought you guys valued windows... and qt5 on windows is a sin... maybe i should force you guys to use wine to compile all my projects to see if you guys would like that :) but now i'm starting to think that this project is based on terrible OOP designs as no one sane would be using a garbage like QT5... so thats where the laziness of having a simple and clean winapi output for windows... becomes a giant mess... and where it ends requiring 64bit in the end... because it was all misplanned since the begin...

<:kekw:808717074305122316>

his most popular repo is porting dlls from vista to xp so vista+ only programs can run on xp :woag

> (requires SASL/SSL and nick registration at irc.com) sucks i know...

<:kekw:808717074305122316>

odd breed of character

Jxl support in blender when

libjxl compiled to win95 wen?

YAAAAAAAY I'VE GOT IT

congrats! 🎉

https://www.techdirt.com/articles/20211210/12134948095/beware-copyleft-trolls.shtml

https://c.tenor.com/ZbmZDT16_FMAAAAM/merry-christmas-images-christmas.gif

Or if you prefer, Merry Xmarx!

I wonder what compression scheme this is

Maybe Microsoft Video 1 or MS RLE?

ah, could be

(sounds awful)

from _Crypto Dictionary_ by Jean-Philippe Aumasson

Whenever I try to set this image as wallpaper, gnome-shell hangs and crashes. I cannot even login afterwards.

I'm going to be a godfather for the second time

Got this present from the still-to-be-born baby 🤣

It says "do you want to be my godfather?"

Printed on bottles of Chimay Bleu, one of my favorite beers, haha

Hm, guess I'll try one of those today <a:ThinkingSphere:821038590091329557>

Hm, that image actually looks nice AI enhanced

~~Also some PS auto settings <:KekDog:884736660376535040>~~

Hm, turns out they're kinda nice

Rochefort 10 is nicer in my opinion, but if I want something lighter then I guess that's my favorite. It's way nicer than expected and might be my default go to beer, since it's not that expensive compared to Rochefort 10 (which costs 3 euros per bottle...)

Chimay Blue is one of my default go to beers in autumn/winter

Rochefort 10 is also nice but indeed a bit expensive and heavy on the alcohol

Understandable. It's nice ~~drank too much though, so my English is probably messed up <:KekDog:884736660376535040>~~

And yeah. Wish it were way cheaper since it's very nice

Chimay also has 75cl bottles with grand reserve versions of the bleu

It ages well, if you can resist drinking it

4-5 year old Chimay Bleu is quite different (and better) compared to a recent one

Also the small bottles age well

My favorites are: Summer: Tripel Karmeliet Other seasons: Chimay (tipsy/drinking in), Rochefort 10 or Achel 8/9.5 when I'm tipsy/drunk

Tripel Karmeliet is also one of my spring/summer favorites

Hm, I have 33CL Chimay bottles here. Second favorite for sure

Nice

La Chouffe is also nice in summer

Oof, typing is kinda becoming a little hard

I visited Achel last week

Nice, I bought La Chouffe and I was actually planning on trying it this evening <:Poggers:805392625934663710>

Like, I bought a big bottle and was going to try it now for the first time in my life <:KekDog:884736660376535040>

(lots of Dutch people in Achel, it was the weekend the Dutch lockdown started haha)

"de hollanders hebben alle appeltaart opgegeten" said the waiter in the brasserie in the Achel brewery

Lol. Many Dutch people seem to be going to Belgium, because rules here are kinda too extreme. The government decided a full lockdown (everything non essential being closed and essential things closing early) would be the best thing

Lol

Their pie was nice, however, I went for cherry when I went there :p

What was funny to me though, they called their "elderberry cider" very expensive and "luxurious", because it costs 4 bucks, but that's almost the normal price of a cheap bear at bars here. Usually a cheap tap in a plastic cup costs 3 euros here... yay city life <:KekDog:884736660376535040>

~~Elderberry cider tastes 10 times better though~~

Hm, La Chouffe is kinda nice

Westvleteren is the hardest to get of the trappists - they make you jump through hoops if you try to buy it

But it's not very expensive - a bit less than 2 euro per 33cl bottle for the Westvleteren 12

Hm, can't say I ever heard of it

Lol my alcohol tolerance. Haven't had a drop of alcohol for 3 months or so, so now even a somewhat moderate amount feels like... a lot <:KekDog:884736660376535040>

One neat trick to remove banding in images, actually calibrate your monitor properly

I didn't realise how out of whack mine was, it's a night and day difference

It's actually quite noticeable with JXL images, really dark sections are actually really dark now, I could see them before and it looked "muddy"

the vcgt does make a significant difference even outside of color managed applications

unfortunately just like everything else color management, there's not a lot of literature on the topic

<http://www.lagom.nl/lcd-test/black.php> and the white saturation one is useful without a colorimeter

I got a second hand colorimeter, basically hands off

Made me realise I had my monitor settings entirely wrong, it still looked fine, but that's because I didn't know what I was supposed to be seeing

yeah time for the ultimate test

i1 Display Studio for $180 AUD

iirc it's 99.5% sRGB

Turns out it records it in the profile it creates

the bands and gamma test are probably the only objective tests the human eyes can do

The above is not off-topic enough

This is an appropriate level of off-topicness: https://youtu.be/mU1MhESEYB8

~~Vertical <:ReeCat:827322381349290076>~~

But nice

<@!886264098298413078> How can I uninstall a manually compiled version of your QT jxl plugin?

Delete libqjpegxl.so from the plugins/imageformats/ folder where it is installed. And also there could be jxl.desktop file in /usr/share/kservices5/qimageioplugins/ . jxl.desktop could be installed also by imageformats package.

Thanks 👍

Sometimes mozjpeg q1 can produce nice looking results

q2

q3

Surprised how decent avif looks. Could probably even be served

Lol, webp just dies at that point <:KekDog:884736660376535040>

<@!219525188818042881> lol i didn't even realise the background was supposed to be a gradient until i saw the webp

why is there such a big colour shift on webp

don't kill me but i think q1 looks better than q2 and q3, cuz it doesn't make any attempt at trying to do the background <:kekw:808717074305122316>

Lol. It looks like it's an actual style <:KekDog:884736660376535040>

Also, the avif actually looks good

<:banding:804346788982030337>

It's not even too bad, especially not for 61KB... 4000x4000 <:KekDog:884736660376535040>

fair

and even tho the lines against the background are being blurred the ones inside aren't being blurred to a really bad level

oh wait i forgot firefox enabled avif support <:kekw:808717074305122316>

how does png and jxl fare?

well if you can get the png down to these sizes <:kekw:808717074305122316>

Optimized PNG is like... 5.68MB and guess it'll look oof'd when made very small

just make it black and white™️

Avif looks way better than jxl tbh, but need jxl q 1, so that was already expected

Hm, e9 q2 might also be small enough <:KekDog:884736660376535040>

Hm, q3

``-e 9 -q 3 -I 1 -E 3 -g 3``

also looks great lol

Of course png looks absolutely terrible at those sizes...

what does the q1 look like?

oof jxl does a lot worse than avif when u zoom in tho

q1 looks almost the same

Well, yeah XD

153KB optimized PNG

but less banding in background <:HaDog:805390049033191445>

but now u get smoothing and a colour shift around the edges

oof

nightmare fuel <:monkaMega:809252622900789269>

so it's actually nightmare eyes with a possessed fox and not a happy fox

Got a smaller PNG if you want <:KekDog:884736660376535040>

65.3KB

<:YEP:808828808127971399>

i'm sending this to foxtopic

Lol

Want an even smaller one with only 2 colors?

34.2KB

64 colors bonus fox <:KekDog:884736660376535040>

what is the original?

smh 1MB

too beeeeeeeeeeeeeg

the dithering u applied there makes a really bad moire effect on my monitor <:kekw:808717074305122316>

Delta palette might work well on this

How big is cjxl --lossy-palette --palette 0 on this one?

We should make variants of lossy palette with more loss

Also lossy encoding of gradient predictor residuals (like what we do with DC) would make a nice near-lossless mode that avoids banding and DCT artifacts while still being a lot smaller than full lossless

Using that didn't change the file size for me ~~no idea how to properly use most parameters tbh <:KekDog:884736660376535040>~~

I mean with -m too

And maybe -P 0 or something is also a good idea

Hm, prefer the image without those parameters. It created blocky banding at -d 1

Is it from a svg?

No

https://www.teeturtle.com/products/boba-fox?variant=19577791455200

> --separate=1 🤔

Cursed

I like pearl milk tea 🙂

🧋

look great 🙂 > jpg q99 444 386,119 > jxl 76,562 > -j -m -s 9 --lossy-palette --palette=0 -g 2 -E 3 -I 1

-lossy-palette --palette=0

Lee jxl got quality improvemrnts

https://artifacts.lucaversari.it/libjxl/libjxl/2021-12-28T23%3A30%3A12Z_7bbbe968fc6a512f9d1fe722abaf10a249566450/

is it softe at high distances especially in infaview

Can't say it looks better to me than without ``-lossy-palette --palette=0``

playes don't eally suppot peopley colos

Just ``-e 9 -q 92 -I 1 -E 3 -g 3`` 77,716 bytes

Don't like the blocky banding ``-lossy-palette --palette=0`` sadly, because it looks very obvious in the background

is it vardct?

Think so

Hm, ``-j -m -s 9 --lossy-palette --palette=0 -g 2 -E 3 -I 1`` looks very bad on a higher quality version of the image

On this one

That's great 🙂 , Could you teach me which PR improve quality? look like PR403 still haven't merge. > https://github.com/libjxl/libjxl/pull/403

No pr

I don't know if modular also imrpoved

But There is better efficiency

See the av1 Discord

You are in it

ok

9a74 is written

yeah, that banding isn't good, I think lossy-palette still very experiment for now, and I look forward jxl separate option can fix everything. 🙂 > https://github.com/libjxl/libjxl/pull/466

Fixed my issue, which was both your plugin and kimageformats providing jpeg-xl

With the image i used s 8 and s 9 d 1 werent heavier abd werent high quality

For quality like you need an image that dont weight 60 kb use d 0.2 s 9

With s 7 It doesnt chahge anything you don't really need full effort

Like in svt av1 p5 film grain 22 input depth 10 crf 16 closed gop 1

Is enough

With the new build

And motions looks good

I downloaded a pipeline from 20:00 29th december

I Guess is some thing veluca is correcting

But you need to wait April 2022 2nd edition of the First part of the specification

18181-1

For audio use 172 kbps cbr fast coder ffmpeg aac

180 abr nero

MP3 v1

MP3 v0

MP3 v2

FDK 1.0.2 VBR5

even gxlame v 27.3 is acceptable

i would like a new apple aac encode3 at p4 160 kbps

o even new exhale p 3 b 62

https://mainichi.jp/english/articles/20211227/p2a/00m/0na/072000c what

based japs resist change

it is at least more secure exploits wise - flash can pretend to be any kind of device, and various usb drivers have been known to have exploitable bugs

it's especially relevant if they are running outdated systems too

chad

https://youtu.be/M4IMiVJBnys

Getting ready for Moscow timezone New Year with the kids — CET is a bit too late for them 😅

<:galaxybrain:821831336372338729>

Happy New Year!

(I wrote it down for my daughter, she added the title)

<:Coffin2:770327766813900870> <:CoffinDance:778934247469613087>

where can i find fjxl?

Scope had a version somewhere in the benchmarks channel

I think

<@111445179587624960>

ah i see, i can probably build it from there

i was looking for a separate repo

Intel = rekt again?

waiting for fan translation

https://www.youtube.com/watch?v=J1Np4y4qc0Q

apparently someone on /a/ already translated it, it's just not typeset yet

nvm it's typeset, at ||<https://boards.4channel.org/a/thread/232541480/ayakashi-triangle-chapter-74#p232546207>||

https://newatlas.com/photography/717-gigapixel-rembrandt-night-watch/

https://rijks-micrio.azureedge.net/ERwIn/20/450_377.jpeg < You wouldn't download a Rembrandt piece by piece and stich it back together

Is there a way to encode an image that doesn't all fit in memory with the current encoder?

Nope

Well technically you could split it in tiles and encode them as separate frames, but that is not how it should be done

Will not be for 1.0 of the api to have encode options that are not full-frame though

But could certainly be added, e.g. encoding 2048x2048 pixels at a time wouldn't be too hard to do and could even be done with streaming output too (no need to buffer the compressed bitstream)

in theory even DC can be encoded in tiles, right?

Yes, DC gets encoded in tiles of 256x256, which corresponds to 2048x2048 pixels

AC is encoded in tiles of 256x256 pixels, so you could also encode in smaller chunks but then you need to buffer the DC in the encoder

Also all auxiliary images like chroma from luma coeffs, adaptive quant weights, filter strengths are also encoded with the DC groups in regions corresponding to 2048x2048 pixels (but those are typically smaller amounts of data, e.g. CfL is 1:64 so only 32x32 values per DC group, quantweights and block type selection are per block so at most 256x256 values but typically less — it could be just 8x8 values if it's all DCT256x256 blocks.

if data can be fed in 2048x2048 groups, we could potentially have quite reasonably low memory use encoder, without too much complication. well, as long as 2048x2048 is 'low' for particular use case

and even DC buffering option doesn't really sound all that bad

though i suspect DC buffering isnt really avoidable -- you want to put it complete in particular place in bitstrean

Of course density suffers a bit if you encode image chunks independently, i.e. cannot globally optimize histograms, cannot do patches across chunks, etc

either something needs to be buffered somewhere, or these 2048x2048 tiles need to be passed in twice

You don't _have_ to put all DC before all AC, you can permute groups so you mix them, at the cost of course of losing whole image progressive previews

makes sense

in practice i think that first writing to temporary file and then rewriting to final one shouldn't be a bad thing

unless you wanna stream output and dont have much temporary storage

Yes, or just keep the compressed AC in memory until all DC has been emitted. Or a two pass approach. Or permute and don't have network progressive (you can still do progressive decode if you have the file on disk, just need to seek)

i think you could even optimize histograms globally if you do enough of passes

It shouldn't be too hard to make a tool that rearranges the groups afterwards, it's just byte shuffling (only a bit annoying that you need jxl entropy decode to interpret the toc permutation)

Yes, if you don't care too much about speed you can just encode twice, basically.

patches across groups sound complicated even if you encode twice if you wanna conserve memory, though

Yes, I suppose best you can do relatively easily is a two pass approach where you find patches per encode tile and remember them all, remove duplicates, and then see which patches can be used. Still won't find repetition where the patch occurs only once per group, but at least if e.g. the letter "e" occurs twice in one group, all other groups can use the same letter "e" in the second pass even if they have it only once.

But I suspect that for use cases where memory is enough of an issue to make chunked encode needed, you don't want to do fancy stuff like patches

3 exabytes of ram

<:Poggers:805392625934663710>

It can, but you'll need a huge amount of memory or swap space...

Also making a viewer for images that large that doesn't run out of memory is quite nontrivial

While we made sure that the bitstream itself makes it possible to deal with huge images (by having ways to do chunked encoding and cropped decoding), for now we haven't tried to implement anything like that yet — we still kind of assume that there's enough memory for the whole uncompressed image (plus some overhead).

But even if we don't do anything new in terms of implementation, assuming typical available memory will continue growing like it did in the past, the maximum _practical_ image dimensions will keep growing anyway

So that upper bound (one gigapixel by one gigapixel, so one exapixel) is effectively just saying "there are no hardcoded limits on the dimensions", unlike e.g. the old JPEG which has 16-bit dimensions so at most 65535x65535, or WebP which has 14-bit dimensions so it cannot do, say, three 8K images horizontally next to each other.

In particular, in some cases you may want to have images with extreme aspect ratios for some reason, say 100x100,000, which is only 10 megapixels so not really an issue in terms of memory, but neither JPEG nor WebP can represent it.

you also need to allow for downsized decoding - either decode only amount needed to fulfill certain resolution, or downsize after decoding tile by tile

this is for example allowed by jpeg, exposed in imagemagick, and useful for cheap thumbnail generation of whatever size inages

> _P_ values have always had critics. In their almost nine decades of existence, they have been likened to mosquitoes (annoying and impossible to swat away), the emperor’s new clothes (fraught with obvious problems that everyone ignores) and the tool of a “sterile intellectual rake” who ravishes science but leaves it with no progeny³. One researcher suggested rechristening the methodology “statistical hypothesis inference testing”³, presumably for the acronym it would yield.

oof

lol

They have a place, but only after a lot of other factors have been controlled for.

do they, though

Lol, someone put a USB flash drive here with Polish music on it... I can borrow the flash drive, so I can update my monitor firmware, but then I have to remove the music first. Thing is, it's write protected and regedit and diskpart can't remove the protection, so guess no firmware update for me <:KekDog:884736660376535040>

how good is the music?

maybe it will make for a good consolation

Not really my taste XD

Also, Windows Media Audio

LCD monitor. Firmware update improved HDR performance, so could be an important update