JPEG XL

no ram error

software does not need documentation that says if you try to do stupid things with it then it might not work

yes, I see that this image is large, but can I find out where the Limit is written, and what is it based on?

I found it *immediately* from google

`((Width * 8) + 1024)*(Height + 128) < INT_MAX`

just because you are bad at doing your own research doesn't mean that the software sucks

I have friends who have at least 128 gigabytes of RAM. but the software just doesn't work.

what? doesn't that make sense? the limit is not shown here

INT_MAX is 2147483647

in pixels is 2147 megapixel

it's not pixels tho

the LHS of this expression isn't measuring pixels

your image has fewer than 2147 megapixels but it fails the size limit due to whatever that is calculating (i guess something with rgba 16-bit channels and some padding?)

that would be my best guess

as this ends up being `8 * (width + 128) * (height + 128)`

and 16-bit rgba is 8 bytes per pixel

oh it does lol

even without the padding, in 16-bit RGBA your image dimensions take up `3136607544` bytes, which is more than `2^32 - 1`

wonder why it's written out that way in code

I don't know, I'm guessing gyan doshi copied it directly from the check

when answering the question

((1091 * 8) + 1024)*(359373 + 128) = 3505853752

``` 3505853752 > 2147483647```

yup

it's measuring bytes required

might be related to GPU code which could still potentially have 32-bit issues

> On a machine with 32-bit integers

which is basically every machine

`sizeof(int) == 4` is true on every architecture in the last 20 years

But anyway, just look at them. they just don't solve this problem <:PepeOK:805388754545934396>

so yes. in my practice, everything worked out with the theory - can't work with the image in my cases more than 1280x190000

I completely disagree with you

Well, if the program does not work because it has limitations in the code of work in a maximum of 32 bit values - of course software sucks

Even the dumb Paint-net has a block chunks structure allow you to work with all these images, even though it can't JUST save a Png with a size larger than 65535

although it is worth mentioning that paint-net has a size limit, this is it, but it has it over the entire height+width (I just don't have enough memory to create such canvas), which means that the maximum coordinates are 36 bits

but this no kidding, it's literally I can work with an image, it takes only 1.5 GB RAM and I can work with it - it's not a problem at all.

jxl could if decoder was sane not use that much ram

but like actually implementing all the details, especially extra features like patches, splines and filters with awareness of current view field and zoom factor is an effort

afaik vardct provides sort of progressive steps, and even in case of modular you could use tiles, resizing each one before store & reusing space for next one

but then applying patches and filtering the same way as you'd get doing filtering and resizing stuff only then could be tricky

it's like, kinda a hack, to not actually decode stuff fullres for display, but in considerable amount of cases you cant really display fullres image if its large, resorting to either resize or part of it

so in other words, jxl can but it cant

png has decode by row and interlacing

jpeg has blocks

it can but its not exactly easy

jxl has nothing

right now

png cant if img width is limiting factor

actually it can is a bad word

it has the potential

but it can't

fair. id say more potential than png and jpg. historically formats didn't have much potential for this, so its kinda under-explored territory in general imo

i think the way avif is doing tiling would allow it too

except i wouldnt want to live in anticipation forever

for editing people end up just using uncompressed

i wanted to do a demo at some point but like this winter is too hard to do anything than bare minimum. i kinda see how but no energy

i don't think it's worth the effort

the parts that make the codec are too hard individually. it would be helpful if those parts were as easy as importing a library

you probably are right. i doubt it'd go further than being a demo even if done, a lot of software is just not designed to take advantage of something like this

compared to creating a png encoder and decoder, which can be done in a matter of minutes

png-simple like format but with tiles would be.. interesting

it does not consume memory, takes 2 gigabytes to display the finished image

When jxl is read sequentially in the case of progressive decoding, it is decoded with macroblocks whose size is the maximum DCT block, i.e. 256x256 pixels

Fourier has many nice properties but one rather nasty property that makes it usually worse than dct -- it assumes an infinite loop of the signal, whereas dct assumes the signal to be just a sample from the middle

I just wanted to show that there were a lot of "ideas" themselves. But in practice, only "dct" is used for all image codecs (excluding jpeg2000) and "mdct" for all audio codecs (excluding speech codecs)

Of course, I'm sure that someone is doing research study somewhere. But is it really that bad?...

Real meme

at a time with pik's development we tried log-gabor transforms

they look very organic, they detoriated towards old photography look

but are slow to compute

and also not that effective at high quality since you need some overlap

https://sciencebasedmedicine.org/an-sbm-advocate-goes-to-washington/

a good article on so-called “integrative” medicine

How to print all value of x265-params in ffmpeg ?

`x265 --help`

they're the same parameters

Okay. it's not a problem anymore. it's just that the latest ffmpeg builds are not detecting hist-threshold parameters. I have a different problem at the moment. SceneCut does not work. at all. for the last 3 hours i cannot understand why it does not work. on any parameters

<@853026420792360980> https://cdn.discordapp.com/attachments/587033245061873759/1093873552287092766/OUTPUT_edit.mp4

I absolutely cannot get HEVC to create a keyframe here. I've read all the documentation. I've looked at the source code. I've asked around. but I can't figure out how to do it.

<@853026420792360980> Usually logic works like this - find parameters at which codec will make the keyframe by mistake on every frame in general - then go to decrease this parameter --- but I have not found such a value! codec at any values does not behave in such a way that I can use it

I can set the value to 2147483647 but it just doesn't work the way it's written

maybe x265 has a bug in its scenecut detection? idk ¯\_(ツ)_/¯

I'm not familiar with x265

~Six months?

what?

it does not work on either the old or the new ffmpeg

well it would depend on the version of libx265 linked in

it also does not work on ffmpeg from 3 years ago

I know AV1 supposedly has scene transitions disabled in the current ffmpeg, maybe it's just a feature that's never quite worked right for years

Scene detection does not depend on FFMPEG, x265 codec does it itself

perhaps it's been a long-standing bug in x265?

idk

FFmpeg doesn't handle the scene detection, x265 does

the thing is, what I'm doing is compressing the game, all the compression is literally based on scene change detection. this allows it to be compressed a hundred times

this

okay?

🙂

maybe it just doesnt want to put a scenecut there

What does it want? IT has to, I have to make it!

it doesnt have to

so it wont

???

😡

I need to check ffmpeg from 10 years ago

Okay..... need to find the archives

https://www.videohelp.com/software/ffmpeg/old-versions

I tried HEVC six years ago and it's exactly the same - it doesn't work

use a different encoder then?

What encoder has a keyframe detection setting? ahahah. maybe VP9 ahahah

I know ffmpeg doesn't handle the detection, I just find it interesting neither of them seem to work

kvazaar has scenecut detection if you want it to be HEVC

but why are you so insistent that there must be an I-frame at that specific point?

and why can't you just lower the threshold until it puts one there

The whole compression principle is based on the fact that the key frame will have a QP of 25 while the P/B frames are 35-40

wait

bad sample

I'm not going to watch it anyway

ok

https://imageoptim.com/color-profiles.html

Absolutely insane that this is their reasoning

ICC is bloat lol

>However, ImageOptim does not convert image pixels to the sRGB color space, because that's a slightly lossy operation. loll

apparently removing color profile which is critical to display image correctly is not a lossy operation...

still many people with TN and VA garbage that can hardly cover at least some of srgb and surrealistic oleds too

that message was kinda half true half repeating message of the article in salty way. i like writing ambiguous stuff which can be interpreted in either direction & still not be wrong :>

the real annoying point was not converting to sRGB on ICC strip

like either strip ICC & convert to sRGB, or don't do either. doing only ICC stripping results in pretty much always broken files & you need original file to know what ICC profile was used to recover stuff

so it's definitely lossy

I think the article I linked accurately explains ImageOptim's reasoning. I just disagree with it

It’s not an unreasonable take. Most users don’t have screens that can display color profiles accurately, and there is software that doesn’t handle them well. The argument is that using color profiles should be avoided because there are programs that don’t handle them well, and few users would be benefitted from using them.

You might argue that “the problem is with the programs” rather than with the people wanting to use color profiles. And I don’t disagree, but the reality is that, as unideal as it might be, the problem is there, and it’s not trivial to solve (as a social problem, i.e. making it well known and getting people to care). The premise, I think, is to live with it and accept it by working around it, rather than to hope people and programs will change to cater to your needs (even if your needs are not unreasonable).

I feel like generally people won’t be so passionate about solving mindset problems unless they are moral problems (e.g. GNU with free software).

Stripping without conversion is not an option for CMYK, and for RGB spaces it only gives somewhat reasonable results if the space is not too different from sRGB. For Adobe 1998 or Display P3, you can do that, and it will basically mostly just desaturate the image. For something like ProPhoto or HDR spaces (or XYB), just stripping without conversion is horrible.

Why not just make it a *super* accessible option, disabled by default, & inform the user? It is a Mac app after all, & every Mac sold right now has a P3 display

Stripping without conversion is never a good idea imo. If you're worried about conversion being lossy, then that sounds strange to me, because the alternative is way more lossy than the quantization error of going from one integer space to another.

Also, isn't every iPhone photo taken now HDR by default?

I don't have an iPhone so can't check

I don't think they produce 'real' hdr images in the sense of RGB pixels with a hdr tf. I think they do the 'backwards compatible' thing where the main image is a tone mapped sdr image, and then there's an additional channel with scaling factors that you can use to reconstruct a hdr image.

Basically like RGBE, if I understand correctly

I also don’t think stripping is a good solution at all. But at the same time, an argument could be made that you can restore the color profile on your own. Not a good argument if the resulting image is user‐facing, though.

That’s pretty cool. Is there any real downside to doing this as compared to having ‘real’ hdr?

Yes. It's less precise. In 'real' HDR (with sufficient bitdepth) the RGB components are independent, while in RGBE they share the exponent/scaling factor. So if one component is very bright but another isn't, you have very little precision left in the darker component.

For delivery of final images it's probably ok, but if you still want to do editing, I think 'real' HDR (say in fp16 or fp32) is way better than 'backwards compatible' HDR in 8-bit RGBE.

Gotcha. & 8-bit HDR isn’t really supposed to be a thing either way right? Isn’t it supposed to be higher bit depth bc the extra color values are more important with higher dynamic range?

8-bit RGB is certainly not enough for HDR. With 8-bit RGBE you can get a bit further, but the main advantage I guess is that you can make it fallback gracefully to SDR and legacy software. Though that also means that in practice, the E will often be just ignored and it's essentially just an SDR image in practice.

Are there any significant issues with real HDR that pushed Apple toward this solution? Couldn’t they have something tonemap & transcode their photos down to SDR equivalents instead, as an option when sending/exporting them? Also does Apple’s use of HEIC really impact their images’ quality per bit meaningfully?

Sorry, lots of questions <:SadOrange:806131742636507177>

I don't know anything about Apple's decision processes or rationale. My guess is they went for hevc hardware (which is useful for video anyway, and I suppose the royalties are something Apple can handle, either through cross-licensing that makes it cheap/free for them, or by just charging their customers for it), and then tried to make the best use of it they could for still images too. Which means doing things in tiles and in 8-bit 4:2:0, because that's all the hardware can handle (at least the common denominator of the hardware they have deployed and still want to support). And then doing HDR separately makes sense, and also maps nicely to the "HDR-JPEG" approach (JPEG XT) which also does it that way. Just guessing though.

> The majority of typical consumer monitors have a color profile similar to the standard sRGB profile, and can't display any “better” profiles. not true, Display-P3 is more and more common, notably on phones

and keeping the *image’s* color profile (or at least converting accurately to sRGB) benefits all screens, sRGB or wider

Huh, interesting. The fact that they use HEIC at all in a world where it is so unpopular is fascinating to me because there must be a benefit to that cost considering nothing supports it. It'd be interesting to see a smartphone put in the effort to spit out higher fidelity, 'real' HDR photos, even if their image quality "isn't worthy" (or medium fidelity to save space)

And yes, heic is bad for quality, but then again we are talking phone cameras here, which are doing tons of signal processing anyway so pixels are heavily processed anyway. The smoothing/smearing effect of heic may even be considered beneficial rather than problematic in their overall use case, and they probably measure things in terms of PSNR where heic always looks very good, so they might be fooling themselves into thinking that heic can really reduce bpp that much compared to jpeg.

I could see AVIF working really well for smartphone photography in that case, considering how smeary smartphone photos tend to be. I shoot RAW on my Pixel because of it

I suppose it may have been true at the time that was written? <@826537092669767691> maybe it's time to update that... Also I don't think I have ever seen icc profiles of 100kb or higher that aren't CMYK ones (which often are 1 mb or so). I am not sure if it's a good idea to use sRGB without icc profile — I would recommend to use tinysrgb or something like that on the web, so you can be sure the image is rendered consistently across browsers. Untagged images are supposed to be rendered as sRGB, but they still aren't always, especially on mobile...

Btw why are we discussing this in <#806898911091753051>, lol

my xiaomi has hdr but it is 50/50 hdr. so it does many photos, combines them and saves to jpeg, so the puny matrix has abit less noises and you dont have white white sky or black trees. but still its jpeg, but with such a hack.

if i import RAW and do some curves thingy, then the sun on it burns the eyes the same way as you would look at it

still these standards arent established. i hope someone decides to promote XDR screens the same way as they promote mega pixels

yes, photoshop allows to set fallback for hdr and thats nice

i think its only for profs / enthusiasts. but some AV company/studio decides to use only one format everywhere?

hdr = (hdr capture =/= hdr format =/= wide gamut)

this would be hdr capture, not hdr format, but not a hack

the subject is too complex for the term hdr that is used too much too broadly

wikipedia page is decent but it doesnt help against confusion <https://en.wikipedia.org/wiki/High_dynamic_range>

lol

reminds me hiend

HD

intel HD graphics

<:trolw:1015162529225392128>

avx-only SHIT

https://github.com/kspalaiologos/bzip3/releases/tag/1.3.0 I really don't understand why IT's not signed as avx-only build. and signed simply as x64 WOW, well, I have x64 and it doesn't work for me - solve your problems

are there any new cpus that don't have avx tho? i would understand avx2 being missing but not avx

Both. but this is not a problem. This is not a topic of discussion because NON-AVX devices are not banned anywhere and there is no "recommendation" not to use these devices anywhere.

I don't know if my phone hits a processing timeout limit or something, but now that it's 5 years old I swear the photos have far less detail than they used to. So I also shoot in RAW every time I care about detail. Has the bonus of correcting overexposure and Telegram even supports direct upload and conversion retaining the 'original' settings

other users is who? does someone have a github?

don't have to. there are no such requirements. I literally don't have to have AVX because.

even the pentium is pretty bare bones and that's not that old <:monkaMega:809252622900789269>

<@456226577798135808>Do you have a github?

i forgot the g4560 has no avx and I also have one <:kekw:808717074305122316>

The old processors are 32 bits. Technically, any 64x processor cannot be considered old. Because it can run any current programs

you just disproved it with this entire discussion <:KekDog:805390049033191445>

it doesn't seem intentional tho

because the program is not x64 compatible

🙏

`"CC=x86_64-w64-mingw32-gcc --host x86_64-w64-mingw32 --enable-static-exe", "gcc-mingw-w64-x86-64"` that's their compile flags i wonder if just changing the march to x86-64-v2 would be enuf to stop the auto vectorisation

It would be correct to write like this - This is not x64 **compatible **program, it should be signed as AVX-only

it doesn't seem like they have any handwritten assembly (well according to github)

it will be hard for me to believe that this person does not know how to compile the program for AVX and for SSE at once - but maybe the new build will be just like that. and then it will be a real x64

I'm still laughing out loud with x86-64-v2 - it's very funny.

<@456226577798135808>you can add this phrase as well:

?

your cpus have SSSE3 and SSE4 right?

does it matter? my chip is x64 - that's enough

what does github actions run now i wonder

just says a 2 core cpu <:kekw:808717074305122316>

Yes. Yes.

well that explains it <:KekDog:805390049033191445>

> ...that said, you're not going to get good performance on these CPUs, since bzip3 leverages ever-so-increasing cache sizes and fast memory. Acceptable single core performance is also a plus because it speeds up the BWT post-coder. The entire point of bzip3 was to work good on modern machines (say, 2016+), maybe using something else might be a good choice for your friend if the performance ends up not being satisfactory.

An acquaintance of mine asked me an hour ago to destroy out bzip3 ☝️ That's exactly what I'm going to do. <:Thonk:805904896879493180> because we both feel that there is something very impractical going on here

In the sense that. Oh, that looks good...... where is brotli 😠

Or, for example, this. here it says (parallel) - is the another codecs in this example also parallel?

Of course it doesn't say how many threads. This looks like manipulation at all. I'm sure in this 👆 example an 8 core bzip3 is slower than a 1 core zstd

it does say wall clock time

wall clock <:FeelsReadingMan:808827102278451241>

which, yes, is unvaluable

*useless

well

<@456226577798135808> bzip3-x86_64.exe & bzip3-i686.exe -- Windows builds don't work at all. They give up files that are 0 bytes in size

This..... looks like a 32 bit build compiled without x64 but with AVX

🙂

I guess you'll have to wait for next release https://github.com/kspalaiologos/bzip3/commit/a7351434f6a44852d40edb0d92edf455d0892e9f

👴

https://tenor.com/view/thanos-snap-gif-25378716

nope

I don't have the faintest idea how to build this on msys2

And I do not understand why the manual says ./configure but the file is called ./configure.ac

you first'd need to run bootstrap.sh, for this you need to have automake & friends installed

I can't understand why this

it's probably easier to go with a package from release, not from git tree, since it already have configure generated

What does that mean? The release package is literally the same as the repo.

try going with one of compressed source releases https://github.com/kspalaiologos/bzip3/releases

no, release has configure script generated

and git tree doesn't

?

hmmm, tar.xz has one

Oh, damn .

this

not the sources.

dumb question: why so many archive formats

linux schozrfrenia

lol

it's compression-geek thingy. For my personal taste it lacks tar.lz 🤪

Only a schizophrenic could have created this 👉 .tar.7z

ok I have no idea what the executable file's dependencies are, so take the entire folder.I have no idea what the executable file's dependencies are, so take the full folder.

now

I have done tests. yes. the format does work better than all existing codecs.... did you think i would say that? it plays back with good old PPMd. bad.

u can go with: `./configure --enable-static-exe` to produce static executable

Great

bzip3 493 225 ppmd 464 660

brotli 614 218 zstd 632 791

LZMA 702 972

it must be said that BZIP**2** loses in general to all these algorithms both in speed and compression

and lzma is not designed to compress text at all

what do you mean by: "it plays back with good old PPMd. bad" ?

why is this bad. I'm a complete ignorant on this.

So that's it. In general we can say that bzip3 is the best at all. That said, PPMd which has been around for 20+ years is better than bzip3 at everything. It's faster. For the same memory consumption it compresses 5-10% stronger, you can even make 1 gigabyte of memory. But it also compresses better at lower consumption. Minus as I think one, PPMd as simitric algorithm takes a lot of time and as much memory to uncompress. Which may be the only plus of bzip3 I can see............... ahahahahaha bzip3 Also simitric and it takes the same amount of time to decompress ahahaha. Well of course I laughed because NEW ANOTHER algorithm is no better than the existing one. I am NOT satisfied.

In general the whole post is built on munipulations, there is no PPMD or BROTLI in bzip3 tests on githab. It talks about paralleling while other algorithms are shown in one thread. This is all very bad.

And of course showing in text compression tests an algorithm that is not designed for text compression - LZMA - is very, very manipulative

I meant to say that bzip3 loses

PPMd is best compress

damn, decompress speeds are indeed a bummer

so much complicated linux thing

it sounds like linux.exe

msys2 thingery?

no, just this program configure thingery to statically link libbz3 into exe, and not load it from .dll/.so

what I mean it isn't universal across different packages, only specific bzip3 configure option

based on the results of the tests and discussion, I can say that the much better general purpose algorithm is still ZSTD

why not to just use ZIP

Zstandard has a better ratio than deflate, but for distributing a bundle like this you really only need zip or targz or something like that

it died

?

i use it

not a great chance zip will die soon

zip is pretty much the only archive format that just works out of the box on any platform

I use zip exclusively for compatibility with anything when my friend is on linux or vinux or ыnux or wherever

ye, and that is a pretty widespread use case, with windows and mac users too

same

and i use zip to share many files

lol

I don't understand why linux has a TAR culture at all. just use zip for legacy and 7z for lzma-zstd and that's it.

Zip does two things: storing multiple files in one file (like tar) and compressing (like gz)

just because zip doesn't support file permissions. too bad 7z also doesn't.

If you don't need compression (e.g. because the files are already compressed), it's somewhat silly to combine the two tasks

I mean, most compression formats do have a fallback mode to deal with incompressible inputs, but you still end up trying and failing, which makes it a bit slow to create a zip file

Tar is made for tape archiving, so no random access, right?

yep, to list compressed tar contents, you have to decompress entire file, which sucks

Does something exist like tar but with random access, full permissions and other niceties like symlinks?

I'd say "dar", but then it doesn't do solid archives

There are probably ways to just put a whole filesystem into a file, but is there something popular and with some tooling that makes it easy to use?

http://dar.linux.free.fr/

easy-to-use is questionable tho

how good it is that it is.

I don't like it when they forbid me to do something, e.g. to remove drivers when they are run like in manjaro

🙂

And of course I will never understand why I should have permissions on the archives when the file system should do it

well, just because you wanna keep them

7z 💯

7zip is the best, I have not seen any other format that is better

it can replace all existing archive formats

And, of course, the built-in encryption system

bad moment is, and that's why I sometimes use RAR - deleting data of identical files, a very useful thing when you compress a lot of very similar files

Since we no longer use disks or unstable connections, or old hard drives that don't check their errors, creating a recovery file in RAR already seems useless

a friend of mine backed up a project they where working on as a .7z

then it got corrupt

now all that data is gone

lol

to repair .7z files you need to do hex-editing or something similar

.rar does not have this problem

I have no idea where you kept the files if they got corrupted but yes RAR would be more suitable for data paranoid

<:Suicide_revolver:855454994475122718>

I have had multiple friends who have had failing hard drives that silently corrupt data

it's pretty common

i think if .apk is zip then zip is good

i have never heard about this irl, only yt

I have 4 harddisks and I have important data saved on 3 of them

important data? i have none lol

except crappy webserver where my assignments are stored

so if it fails i get F

Wouldn't want to lose 80 hours of music I've been collecting for 4 years ✍️

that's the worst thing can happen to me

If I had been working on a project you can be sure I would have saved it 10 times.

don't melomaniacs have the shells full of CD/s vinyl or tapes

bruh

I'm too smart to be a believer.

lol

opus 128 👍

there's an non zero probability that NASA satellite falls on your house destroying all data including

but backups are still important

don't be me: having 3 laptops, a PC, flash drive, portable hdd, a webserver and all of these have random garbage, and duplicated many times so sometimes i throw away all files i didn't need for a year

No. My data is not that important. it's worth saying that many people overestimate their shit

exactly

I had it happen, only realised it was dying when windows started doing disk repairs on every boot and UI elements were failing

Yeah, the Pixels do HDR+ processing & the iPhones do Smart HDR. It's not really an HDR photo that will display as HDR on the right screen, just multiple exposures layered with processing. Samsung devices do this very aggressively

+

btw i dislike phones because indoors the photos are crappy

outdoors the photos are decent tho

The Pixel maintains the dynamic range multiple exposures provide in the RAW photos. So I just manually edit them to make them look good. I think phone photos look like smeary crap 95℅ of the time

raw on mobile?

its like a balloon instead rocket

ive tried that, noise is incredible + optics issues. all black places are full of green snow

modern phones have cool algorithms that show the user what he loves to look at

My phone's raw retains the slight green tint from it's optics, keeps the exposure perfect with no editing while somehow the normal photo is washed out, and naturally keeps every detail since it's not applying a messy denoiser

i dislike phone photography

selfes / street photo ok

everything else is noisy crap

small sensor -> less light -> increase sensitivity -> more noise -> increase resolution -> pixels are smaller -> less light

“less light” is enough for “more noise” (or let’s say “worse SNR”), no need to involve sensitivity

and smaller pixels means less light per pixel, but if the sensor area is the same, still roughly the same amount of light for a given fraction of the image (with today’s gapless microlenses which result in essentially a 100% fill factor)

(except possibly at very low f-numbers due to pixel vignetting; backside illumination helps with that)

one factor that may negatively affect the performance of smaller sensors is optics: https://www.strollswithmydog.com/equivalence-sharpness-spatial-resolution/

for one thing, diffraction may be a problem sooner

my conclusion is that it's already been decided that phone cameras are *good enough* for most people and when something like megapixels is brought up, it's just bait

-> more megapixels -> more advertising -> ??? -> PROFIT

<a:PepeHackerman:806632585429909514>

works

Are there any people with a ThreadRipper or something who would like to compress a couple of days av1 for fun?

You’ll prob have better luck just using Google colab

is there value in bzip3 -- I looked at it superficially and it seemed more a marketing trick to name it as if it was some sort of next version of bzip2

7zip is a huge collection of stuff -- it is like building a house by choosing that the foundation should have steel pillars, concrete pillars, wooden poles, shallow footing, some deep footing, etc. -- it can be ok to build, but there is going to be some increase in long term cost and some diversity in behavior in operation

there are 2 CPU cores. My 12 year old PC is faster.

for archiving I'm a fan of lzip (and plzip)

how about lrzip?

but you can just leave it running for free

Squashfs images are sort of this.

Unfortunately most linux server images don't actually have unzip preinstalled, so tar does make sense for linux releases.

Tar also has the advantage of because it is compressed as a stream you often get better compression than zip as similar files can be deduplicated.

pxz

.zip.gz

I wrote a script to sort files by extension before tar+gzip to help the lz77 deduplication process

as gzip dictionary is small (32k)

i have a few of my own image conversion scripts, one that i use often is one that converts a png to max quality webp lossless, and keeps all exif data by extracting it to an xmp file with exiftool, and adding it to the webp with webpmux it's kinda dumb that cwebp doesn't have an option to copy exif data normally, and that i have to do this roundabout way

right. What I like about the tar + gz approach though is how it separates the two aspects (a multi-file container and compression), while squashfs and the usual archive formats (zip, 7z, rar etc) are doing both at the same time, which is a less 'modular' approach

but I guess it's tricky to have random read access, easy adding/removing of files, and a modular approach all at the same time...

nope, only ICC profile, seems like it can't copy it on windows platforms at least, unsure about linux, which is odd

right, 7z has the command-line option `-ms=e` to do something similar

(`-ms=on` is all solid, `-ms=off` is not solid, `-ms=e` is solid on a per-extension basis)

I like that 7-zip supports this “hybrid” approach to solid compression, where you can group certain files together compression-wise without having to make the whole archive thus

strictly speaking tar+gzip does too as both can be concatenated

this is what .pxz takes advantage of

can you easily seek to the files with a specific extension without having to decompress more than needed?

don't know about ".pxz", but tarlz does something like that in "--no-solid" mode

(i mean the "enchanted" tarlz <https://www.nongnu.org/lzip/tarlz.html> not simple tar + lzip)

to some degree, yea. it uses some GNU tar extensions like filename entries and takes advantage of XZ's concatenation (like gzip) to manage it

but I'm not sure all of the implementation details

I just know it recognizes itself but it can be treated as a .tar.xz

It turns out there's a good way to display 18,000 image previews with links to them, and it's HTML only, is a really cool thing <map>, and I wonder why it's not used more often nowadays...

because 18,000 previews with a <map> requires one image with all 18000 previews, which is a really bad idea

It doesn't fit in one file. Gif has a limit of 65535, jpeg too. And png larger than 65535 is simply not displayed in firefox, it is literally banned

Therefore, I use images of 15x15 images of 80x pieces on the site

80x15x15= 18000

`<map>` is quite neat from what I remember reading about it. I remember that it allows link areas to have distinctive shapes, even!

Waterfox is AVX-only? lol

it's even worse.

and again, the phrase x64 compatible has nothing to do with real x64 processors.

But they didn't write literally anything. Generally. -- This software for Windows - Well, Windows works for me, but YOUR software does not.

fuck them

i see it

I remind that to run Windows, the minimum requirements of the SSE2 instructions (2000). And the All Windows system with all its drivers will work

I didn't have any problems with running any Linux system either

Is this inherent to waterfox or is it just a limitation of the Windows build/installer they offer?

Judging by the comments of the developers it is literally hardcoded

I can't make a build of firefox, it's not enough to live time...

What is your goal, ultimately? I suppose you want a “modern enough” Firefox fork that supports JPEG XL, is that right?

yes

hmm, have you tried Librewolf? I don't use, but they merged this: https://gitlab.com/librewolf-community/browser/source/-/merge_requests/47

well. no, the latest librewolf-112.0-1 does not support Jpeg XL

Linux build for 111.0 seems support it, after changing `image.jxl.enabled` to `true` in about:config

lol

👆 Is that a discrimination

there are literally two browsers in the world with out-of-the-box support JXL: Thorium and Palemoon

Sadly, looks librewolf relies solely on firefox nightly code, which doesn't support animated jxls

You should be able to change it by double‐clicking on the “false”.

problem is it's not "out-of-the-box support"

It's not a problem. the problem is that if I say that librewolf can be used to view Jpeg XL, it will be a lie. - The user will download the browser, open the image, And won't see him. It's disgusting.

I don’t think it would be a lie. I understand it’s not as convenient or straight forward, but you can definitely use the program to view JPEG XL images.

In the sense that viewing JPEG XL images can be done by using the browser.

so can chrome, if you patch it yourself

the thing is. what should you answer me to the question - why did the developers turn off jpeg xl by default?

You could argue it doesn’t “support JPEG XL images”, but not that it can’t be used for viewing them.

For me, it is quite equivalent to the fact that this software is turned off = it is not there at all. Because in the same way I can put the jpeg xl extension on any other browser - for this it will be necessary to do some business - this devalues the fact that Jpeg XL was added to librewolf in general 🥱

It's default firefox-nightly setting, so they just didn't change it. Also as they're privacy-focused I don't think they will change that unless firefox changes that, because of browser fingerprinting

i agree with dzgas here

the browser can't view jxl

the browser with the setting can view jxl

I say that “the browser with the setting” is still “the browser”. What you’re saying is just like saying that a website doesn’t have a dark theme because it isn’t the default.

saying dark "theme" is like saying dark "setting"

it's not a dark browser

it's a browser with a dark setting

topic 👉 🥱

but that value is different to different people

Instead of recommending librewolf, I will say this: there is **palemoon**, there is **Thorium**, and there is an **extension **that you can install on any of your browsers.

(Which I’m working on updating right now, by the way! 😄)

ye, in terms of jxl support librewolf can be only recommended to "power users", not general public

it has libre in it so it must be bad

I will say: I think I understand what you mean now. Where the setting is in effect a change to the user agent itself (as in, it changes how the user agent behaves in the eyes of the user).

users understand changing a theme, a setting designed to be changed

users dont understand changing a developer flag

That’s reasonable. I suppose it’s a matter of how the setting is presented, then? If it were on the settings page of the browser, would that change your opinion?

no because nobody knows what the hell a jxl setting is

I see.

to me it just sounds like the browser doesnt *support* it

it supports the capability of it

but it doesnt actually support the usage of it

lol, reminds me of this: https://youtu.be/jmaUIyvy8E8

user tries jxl doesnt worK

moves on

I just don’t think it’s fair to say that they browser can’t display JPEG XL images. I think it’s fine it doesn’t support them, in the sense that it doesn’t encourage or endorse their use.

would you say that windows photo viewer can display jxl

I don’t know. But I suppose your point is that it needs an additional user action (be it changing the setting, or be it installing something) to display them.

it can display anything, with a plugin, and to some people, they don't know that, so it just means it can't

I understand.

by that logic, if it doesnt work by default, it's almost as if it won't work at all

Yeah, I see what you mean.

The word "support" is a bit ambiguous in English imo. I would say a browser supports dark mode also if it is in light mode by default and it is a setting you need to change. There is "support" like in "football supporter" (rooting for something / endorsing / trying to push it), and there is "support" like in "support beam" (something functional that is there so you can build on top of it). In software, I think typically "supporting a feature" is meant in the second sense and not really the first sense, though it's probably always a bit of a mix because adding a feature always implies some kind of endorsement.

You definitely don't usually see browsers improving on supported formats, just going "Yeah, we have that now"

https://blog.pkh.me/p/39-improving-color-quantization-heuristics.html

Is “quantization” the same as “dithering”, and isn’t it a reasonably well solved area?

Palette quantization is trickier than simple dithering of high bit depth using lower bit depth

Quantization is "how do I choose a reduced set of colors which could best represent the full set of the original picture". Dithering is a trick of signal processing which consists of moving error from low frequency (banding) to high frequency (grain) by introducing HF noise, or any other more suited method. Because humans prefer it this way (works for audio too)

it’s not just that: it decorrelates the error from the signal

e.g. if the signal is a constant 0.4: without dithering, you would just quantize all samples to 0, so the error would be a constant 0.4: 100% distortion

but if you dither, you get a quantized signal that is 0, 60% of the time, and 1, 40% of the time, so it has the correct average of 0.4

it just fluctuates around it (noise)

https://www.aes.org/e-lib/browse.cfm?elib=4523 > The work was motivated by some common misunderstandings about digital systems. It is commonly believed that small signals or signal details are lost if they are smaller than the quantizing step. Expanding on previous arguments, it is shown that this is not true when the signal to be quantized contains a wide-band noise dither with an amplitude of approximately the step size. The introduction traces the use of dither from video quantization through to its use in audio. quantization error is studied in some detail and the effect of dither is analyzed theoretically and experimentally. By examples of quantized signals it is shown that the dither effectively turns signal distortion into low-level wide-band noise by linearizing the averaged quantizer staircase function, which is as perceived by the ear. https://sjeng.org/ftp/SACD.pdf > All analogue-to-digital and digital-to-analogue conversions and intermediate digital signal processing steps were performed in the linear, multi-bit PCM format, using internal processing wordlengths greater than the desired final numerical precision. One primary benefit of this format is the fact that such systems can be rendered completely linear, with infinite resolution below the least significant bit (LSB), by the adoption of proper dithering at each quantizing, or (in the case of editing and signal processing) at each requantizing, stage. Such dithering, with the optimal triangular probability density function (TPDF) dither, in principle completely eliminates all distortion, noise modulation, and other signal-dependent artefacts, leaving a storage system with a constant, signal-independent, and hence benign noise floor (see [3] and [4]). This is now well understood, and such practices have been the norm in the industry for over a decade.

https://www.strollswithmydog.com/sub-bit-signal/

https://youtu.be/cIQ9IXSUzuM?t=695

<:BlobYay:806132268186861619> this guy

https://www.youtube.com/watch?v=ZVT02Ljnv0U

it's just amazing how few views there are here

Monty is great, sucks he got fired from Mozilla together with the rest of the Daala folks in that big 2020 layoff round. They would probably have made animated avif work in firefox much earlier, and perhaps they could have provided some pushback against the silly political games that led to jxl still not being supported in firefox and chrome. These folks have scientific and ethical integrity, which I always admire.

It sounds sad. And besides, I can't find any of his videos or works for the last 3 years.

i actually spent yesterday messing around with quantization & dithering it was pretty simple for hi-res b&w screentone manga though, a 4-colour palette did really well, was pretty much visually indistinguishable ...too bad the source scan was full of moire

you do have to make the palette yourself though bc the algorithms ive tried tend to do really bad with it, they tend to lighten the black a lot which is especially noticeable since theres large continuous swathes of it

my favorite error message

For modern encoders this should be enough.

96 cores, 1152mb of cache.

perhaps best for encoding many images at the same time

Hmmm. 😅

awww

Many, MANY images at the same time.

Where did you get that? Lol

This won't help with online streaming<:ReeCat:806087208678588437>

>light mode

https://tenor.com/view/reaction-my-eyes-cant-unsee-burn-gif-7225082

Mozilla is very very weird these days. Couple years ago, lots of press, lots of community involvement. Now, very little.

> In 2020, after returning to the position of CEO, her salary had risen to over $3 million (in 2021, her salary rose again to over $5 million[15]). In the same year the Mozilla Corporation laid off approximately 250 employees due to shrinking revenues. (from https://en.wikipedia.org/wiki/Mitchell_Baker)

I don't know anything about the internals, but to me it looks like Mozilla went from something that was primarily a community project to something that is basically just like any other company, with managers doing layoffs so they can give themselves higher wages etc.

the whole thing is just awful

It's amd epyc-9684x. Around 1300$.

Is it yours?

Lol that's so great. Yup just what the world needed. Awesome. Is it just a fact of life that all large bodies where management is deferred tend toward uniformity with the rest of society?

I could afford that if I saved for a few years maybe. If I moved out of my apartment, I could afford it almost instantly (paycheck wise)

Brisette (April 1998 - 18 April 2016)

hello im an telecomunications student of in the politecnic university of madrid, and I have to do a project on the FLIF compression format. My question is it related with JPEG XL, and if so, can someone tell me a brief story of its creation, because I dont find that much info about the subject

impressive nails)

yeah I have read the wikipedia but the information there was very scarce and doesn’t even mention who the creators where and how was the process

so we had to investigate a bit to reach this discord

I stumbled across the PDF from the trials for JPEG XL: https://infoscience.epfl.ch/record/270332 Btw, does someone know what were the five other proposals, besides selected PIK and FUIF?

oh yes sorry didn’t realized

https://docs.google.com/presentation/d/1LlmUR0Uoh4dgT3DjanLjhlXrk_5W2nJBDqDAMbhe8v8/edit#slide=id.gb8ddd4f56d_0_6 here is some info