JPEG XL

what I probably can recommend on this topic is: https://www.photonstophotos.net/Emil%20Martinec/noise.html https://www.dpreview.com/articles/8189925268/what-s-that-noise-shedding-some-light-on-the-sources-of-noise https://doi.org/10.1117/3.725073

also this nice simulation from Wikipedia: https://en.wikipedia.org/wiki/File:Photon-noise.jpg

I think I understand. So effectively there are different factors that combine into three essential sources of noise - randomness in distribution of photons, - high temperature affecting the sensor in some way, - fixed noise, differences in light sensitivity across the sensor. The first one is corrected by receiving more light, but *can* introduce more fixed noise if this is done by taking a longer exposure, but that can then be subtracted out by letting the camera take an exposure with the shutter closed and subtracting those values (or dividing? idk) those values? I have the option to make it do that if the exposure is longer than 1 second

these look interesting, thank you!

What do you mean by "shutter closed"?

also read noise

fixed noise has two components: dark signal non-uniformity, which you can remove by subtracting a stack of dark frames (to average out read noise), and photo response non-uniformity (that’s the one that increases with signal), which is trickier to remove because you need to stack “flat fields” rather than dark frames

(and you divide instead of subtracting)

So to compensate for all fixed noise you would need to take a handful of closed shutter (in)exposures, and also remove the lens, place the camera against a fully diffused source of light, take an exposure and then interpret these values accordingly and subtract them from the photograph you're actually trying to take...

something like that, although you can potentially keep the lens on (and then the flat field will also correct for vignetting)

Could you record the flat field once and re-use the data?

speaking personally, I find that flat fielding sounds like too much trouble to bother with outside of specialised applications

I think that should work

there’s just the practical problem of having software to actually apply the correction

Hm, might consider that for some astrophotography I planned on doing

software for astrophotography stacking might have such a feature

but for daily photography one edits in Lightroom, DxO or the like, it might not be worth the hassle

I think I also saw something like this in a menu in RawTherapee

yeah, if you right click a photo in the file browser it has a 'dark-frame' and 'flat-field'

If it's anything more than "take off the lens and aim the sensor at cloud cover" I don't think I'm going to put in the effort but I'll see about that once we get anything *but* the cloud cover that's demotivating me to take photos <:KekDog:805390049033191445>

I always just think of it as "Shutter speed is light entering the camera, ISO is how much the signal is boosted from the sensor" Usually the shutter speed is set low for responsiveness and to avoid blurring, but then ISO has to compensate. If you lock ISO to the lowest possible, the shutter speed *should* compensate but it will be a very slowly resolving image unless extremely bright

Makes sense

One more thing, does decreasing the saturation capacity mean you effectively reduce the dynamic range of "brightness" the sensor is able to record?

Is that a just an inherent property from turning up the gain, e.g. previously the range was 0-100 light units but now it's 0-10 light units because the ISO multiplies it such that that range is mapped to 0-100, or is it some other property of increasing the ISO that does this

it’s like you describe: the gain affects the analogue range that the ADC can digitise

(https://photonstophotos.net/Charts/RN_e.htm#Canon%20EOS%20R50_14)

I found unsourced claims that the R100 has the same sensor as the R10 so I can use that as a reference in that graph

So read noise goes down as the ISO goes up, or am I reading this graph wrong edit: maybe it's signal to noise ratio? It doesn't say that, but that would make more sense

Changing image profiles normally only affects the OOC JPEG, not the RAW

It wasn't about image profiles, it was about a decrease in dynamic range with an increase of ISO

Ah ok

Yes image quality decreases with higher ISO

In pretty much all regards

Which is why you should definitely spend thousands of dollars on fast primes to always shoot ISO 100 ;)

Luckily in camera image stabilisation has come a long way

Helps immensely

got myself a nice 12mm F2 specifically because F4.5 wasn't cutting it for me :3

Very nice

https://phillipreeve.net/blog/review-laowa-10mm-2-8-af-the-beginning-of-a-new-era/ This lens also looks interesting for ultra wide shots I am actually considering it

my lens doesn't have image stabilization but it's so fast that I can almost always shoot at high enough shutter speeds without introducing motion blur

Does your camera have stabilization?

Unfortunately not

I have the R100 body, i.e. the "how cheap can we make a mirrorless camera body to get people into the Canon ecosystem" camera body

USB 2, thats gotta hurt during transfers

But the sensor is actually really good *and* it doesn't have a low pass filter, which made it a good/fair choice given that I had a tight budget and wanted to dabble in astrophotography (which prefers not having a low pass filter)

Have you considered getting an astro tracker? Can also be DIY'd with a raspi I think

I haven't, but currently my interests are mostly landscape shots where the sky is the main subject, but still having some land in the frame. Which is why I got a fast wide angle lens instead of a fast telephoto lens which you'd otherwise prefer for astro

Wide angle is also nice for astrophotography

Yeah! I've seen some amazing wide angle shots, including with my lens (well, the non-AF version of my lens, but it has the same optical design, except that the AF variant has improved on coatings) It's just that every since I got it the skies have looked like this

https://embed.moe/https://cdn.discordapp.com/attachments/806898911091753051/1333376547238711296/cloudscloudsclousdclouds.jxl?ex=6798ab2d&is=679759ad&hm=6db76a699dbe106ecea5a179ac8e5a67bfdd69f9641418febedbe0dd8138607a&

Oh wow it's a full-frame lens too

With such a wide angle that's very impressive

Looks British

Okay wait, so if that's the case, if you increase the ISO and then normalize for image brightness by reducing the exposure, you're not actually shrinking the dynamic range of brightness you can capture, you're just losing data due to quantization As if taking a 16 bit per channel image and turning it into 8-bit: there you're just mapping the range 0-65535 into 0-255, so your quantization becomes worse, but the effective dynamic range of "min and max brightness you can have in the same shot" stays the same, no?

So with "reduction in dynamic range" we mean "you're no longer using the entire range of values the sensor can provide", but you're not actually reducing the range of brightness values the sensor can produce

Correct me if I'm wrong

(meant to reply to this message)

read noise, projected back to absolute input units, goes down

the noise floor is raised (relative to saturation) and it’s the lower bound of DR

quantisation is not so relevant (camera manufacturers basically use “enough bits to encode the noise”) https://www.strollswithmydog.com/sub-bit-signal/

bit depth is not the main driver of the noise floor (if it were, it would imply an insufficiently dithered system and therefore some amount of distortion)

https://www.reddit.com/r/linux/comments/1ib3vaf/facebook_considers_linux_and_related_topics_a/

LMAO

Lol, this is like when the the NSA labeled Linux Journal readers as extremists -# [(source)](<https://arstechnica.com/information-technology/2014/07/the-nsa-thinks-linux-journal-is-an-extremist-forum/>)

Strong offtop here, but I really don't want to dive into av1 forum. What is the best quality\speed av1 encoder available in ffmpeg? I want to try to get off the h.265

aom-av1 is adequately fast in my opinion

Just use tiling

If the source is 4:2:0 I recommend svt av1 psy

It can be 4:2:0 or 4:4:4, depends

Svt av1 doesn't yet support 4:4:4 so use regular aom-av1 for that

https://wiki.x266.mov/docs/encoders/aomenc#recommendations

There is also a Psy fork of aom-av1 but idk where to get binaries for that

<@401816384109150209> do you provide any binaries for aom-psy101?

I used to, then my laptop with Windows on it was stolen about a year ago, since then I did not anymore...

but I can quickly compile you a Linux binary

And there is this project for easy building on Windows: https://github.com/Uranite/aom-psy101-win-build

Their WhatsApp is the safest claimed by some people

Could you explain this?

Some av1 encoders support using tiling to multithread encoding, improving speed massively in some cases

``` $ ffmpeg -hide_banner -h encoder=libaom-av1 ... -tiles <image_size> E..V....... Tile columns x rows -tile-columns <int> E..V....... Log2 of number of tile columns to use (from -1 to 6) (default -1) -tile-rows <int> E..V....... Log2 of number of tile rows to use (from -1 to 6) (default -1) -row-mt <boolean> E..V....... Enable row based multi-threading (default auto) ... ```

Yes, I saw it, but have no idea what to do with it. Should I calculate it manually?

The tiling that can be used depends on resolution, but try if 4x4 works and adjust down if needed

Thanks

Avifenc has --autotiling now

For a long time... But it's not good, very rudimentary (I say this because my own implementation in aom-psy101 is vastly superior)

https://gitlab.com/damian101/aom-psy101/-/commit/50b9335a25cb1617cf82cf62c162508a303afeba ```c static unsigned int max_tiling_log2(unsigned int sb_units, unsigned int min_sb_units) { unsigned int max_log2 = 0; for (unsigned int log2 = 0; log2 <= 6; log2++) { // For log2 == 1, check all tiles for min SB unit limit if (log2 == 1) { unsigned int tile_width = (sb_units + (1 << log2) - 1) >> log2; unsigned int residual_width = sb_units - tile_width; if (tile_width < min_sb_units || residual_width < min_sb_units) { break; } max_log2 = log2; continue; } // For all other log2, check only non-residual tiles for min SB unit limit unsigned int tile_size = (sb_units + (1 << log2) - 1) >> log2; if (tile_size < min_sb_units) { break; } max_log2 = log2; } return max_log2; } static void set_auto_tiling(TileConfig *const tile_cfg, unsigned int frame_width, unsigned int frame_height, aom_superblock_size_t sb_size, unsigned int auto_tiling) { const unsigned int sb_shift = (sb_size == AOM_SUPERBLOCK_SIZE_64X64) ? MIN_SB_SIZE_LOG2 - 2 : MAX_SB_SIZE_LOG2 - 2; const unsigned int mi_cols = (frame_width + MI_SIZE - 1) >> 2; const unsigned int mi_rows = (frame_height + MI_SIZE - 1) >> 2; const unsigned int sb_cols = (mi_cols + (1 << sb_shift) - 1) >> sb_shift; const unsigned int sb_rows = (mi_rows + (1 << sb_shift) - 1) >> sb_shift; // Minimum number of superblocks per tile dimension unsigned int min_sb_units = 4; switch (auto_tiling) { case 1: min_sb_units = 32; break; case 2: min_sb_units = 24; break; case 3: min_sb_units = 16; break; case 4: min_sb_units = 12; break; case 5: min_sb_units = 8; break; case 6: min_sb_units = 6; break; case 7: min_sb_units = 4; break; default: return; } if (sb_size == AOM_SUPERBLOCK_SIZE_128X128) { min_sb_units >>= 1; } tile_cfg->tile_columns = max_tiling_log2(sb_cols, min_sb_units); tile_cfg->tile_rows = max_tiling_log2(sb_rows, min_sb_units); } ```

lol did the nft profile picture account get banned

i saw them join, thought "i wonder if they'll still be here tomorrow", went to bed, and here we are <:KekDog:805390049033191445>

this is JPEG XL, not JPEG NFT, smh -# [yes that's a thing, don't worry about it](<https://jpeg.org/jpegnft/>)

half considering making jpegnft.info site, and making it an nft-themed satire site in the same style as jpegxl.info

Hello guys, I am going to buy an OLED laptop from asus. One has an 8 bit panel and the other has 10 bit. Both of them have 100% dci-p3. Is there any noticeable difference on day to day usage?

If you often take HDR photographs with the blue sky

10bit can be a massive upgrade in many games

Would 10 bit also reduce banding without dithering in SDR 10-bit media, or are the extra bits just for the extra dynamic range?

oh it gives you 768 more values per channel, so probably also the former

Yeah that's why I mentioned the blue sky where banding is very obvious

you mentioned HDR specifically, so i wasn't sure if 10-bit SDR images would also benefit from less banding on a 10-bit HDR display

im not sure I understand the question

Do those... Exist?

10-bit sdr images? i think so?

you can have higher than 8-bit precision still within the sRGB color space for example, no?

That's still 'HDR' though, isn't it? You have 1024 values instead of 256

lots of 10 bit sdr images exist

i imagined it more like, you're directly mapping the 1024 range to what the previous 256 range represented. so you're not getting a wider range of brightness or color, you're just getting more precision, so more colors you can represent, but still all within the sRGB space. but i didn't actually consider whether this is how that would work

no

HDR refers to the transfer the image uses

that was my understanding too. you can have an 8-bit hdr image

you wouldn't want that, but you could

on linear yes, the more steps the more brightness, but for every image that has a transform applied, the increased range gives more precision, this is important because 256 steps is simply just not enough in most cases

in many games if you keep an eye out, when rendering 8 bit you can still see banding

ah yep that's what I meant

I wonder why games don't just render to like some higher bit depth by default, supposedly your gpu would then dither that down to 8-bit for you depending on your monitor. worst case, you just get 8-bit without dithering like you did before

for some reason higher bit depth dithering doesn't work by default on linux for me and it's not clear to me how to enable it

they kinda do, most games render internally at 16bit, and then truncate down

dithering is actually kinda really distracting in a lot of contexts

some games add dithering, some add "film grain" some do temporal smoothing etc

I know they do internally yeah, so it's more that i'm confused *why* they truncate down. can't they just display 16 bit and let the OS manage it?

but like, a lot don't minecraft for instance just poops

some people use reshade for dithering too

Outer Wilds is the only game I've known to have adjustable dithering, this screenshot was set to max https://discord.com/channels/794206087879852103/794206087879852106/1334871215462154331

Uses Unity internally

thats neat

I wish more games just gave you control over these kinds of options

Noita does too

Noita actually has a lot of interesting graphics settings. It's a game with pixel art graphics, and it has a 'low resolution rendering' mode where the in-game pixels you see are directly map to output pixels, as evidenced by screenshots being very tiny. It also has optional dithering and pixel anti aliasing which can all be controlled

The dithering in Noita isn't adjustable, you can just turn it on or off

I think Voices of the Void also has a dithering toggle, and it uses artistic Bayer matrix dithering

iirc unreal engine defaults to 10 bit sdr output on windows?

basically your only options on windows are 8 bit int, 10 bit int, and 16 bit half float (intended for hdr, not sure if it makes sense to only use the 0-1 range in sdr?)

10 bit sdr is weird on windows because the desktop is only ever composited in 8 bit (unless you turn on auto color management, which i think is the default on some devices now?)

so the only way you're gonna see any of those extra bits is if the application bypasses the compositor, i.e. it uses exclusive fullscreen or gets promoted by directflip or mpo

and then it also depends on what your gpu does with it, e.g. full range 8 bit output on nvidia defaults to dithering off so it'll just truncate it

whereas limited range 8 bit output is dithered by default

oh hmm

i hope that wayland color management proposal is going to get some traction soon

i can't wait to live a life without color banding

noita?

An indie pixel Art game

icic

one again, indie game supremacy

https://youtu.be/0cDkmQ0F0Jw

looks good, Im not massive into pixel game/2ds any more, but i will try it out when I can

it's really fun

It's really hard

definitely noita in top 5 best games i have ever played in my life

I let ChatGPT find a pixel game I played in late 1990s

I gave a brief description and it solved my question after over 20 years

did Discord just change their web UI?

I didn’t remember the server icons on the left being so rectangular, and the text input is now one row for actual input and one row for buttons

this is the current web layout for me still

https://fixvx.com/WumpusCentral/status/1886515032906743981

yeah and now it even has a title bar (even in the webapp) 🙃

broke my Vencord theme...

they added so much nitro related buttons to the message bar that they just gave them their own layer <:galaxybrain:821831336372338729>

But wait, there's more!

I don’t like it

so much wasted space

i use window tiling and i like to have discord as a tiny quadrant sometimes and that already isn't the best experience, so i don't think i'm going to like it either

i suppose i could zoom out though... why didn't i think of that before

This looks awful 😔

Not happening for iOS

true ~~light mode does look awful ~~

Many people use Light Mode at night here

And turn the computer Dark Mode at day time

Light mode helps with text readability during the day.

now thats heresy

I don't mind "light mode" in general, but discord's theme is abysmal

I use catpuccin for theming discord, which android supported that

It's a shame discord's default themes look soo bland

Vaguely on topic, but you can re-enable the devloper tools in the client by adding `"DANGEROUS_ENABLE_DEVTOOLS_ONLY_ENABLE_IF_YOU_KNOW_WHAT_YOURE_DOING": true,` to the `settings.json` in the install directory

Very user friendly

We should recreate this with JXL

https://cdn.discordapp.com/attachments/530853682678595594/1336672175121498142/20250205_121743.jpg?ex=67a4a878&is=67a356f8&hm=5ae34872a3dfae2fe1d12f0296330f806386962c69733e321928e897ee680ca8&

(Yes this particular repost is a little soft)

~8264.65 km^2 if DPI 300

or ~143483.6 km^2 if DPI 72

Looking into this is also how I found out my phone is 400 PPI and my monitor 100 PPI

In the end it depends how you scale things. Since that PDF is purely arbitrary with no definition of DPI or PPI. Even if a JXL were 1 PPI, we'd have most of the planet covered I'd guess

Looks like JXL represented with small LED bulbs

https://youtu.be/RSrRjM0upXY

I like the intricate build-up over the first few minutes

oops, I may have forgotten to specify full range when asking ffmpeg to convert the cover picture from sRGB to BT.709

This reminded me of these cmake flags for Clementine music player. This used to let it build against system libraries instead of the patched versions they included. The Debian packager was not deterred.

is there a way to get some kind of objective metric score for audio using ffmpeg?

can be something simple, not looking for a perceptually very relevant metric

I wonder how accurate it would be to map the audio to an image and compare those

Probably not very

Waveform comparison? I've tried converting to an image before to store audio as a JXL, but there always seems to be some kind of loss

if one doesn’t care at all about visual interpretability (as with a spectrogram), there’s always the option of converting each audio sample to a single 16-bit pixel

with interleaving for multichannel content

either as a very thin and long `nchannels`×`nsamples` image, or directly reshaped to a more usual “aspect ratio” (with separate metadata to know the number of channels)

… or using jxl channels, I guess

there is the apsnr, asdr, and asisdr filters

That's kinda what I did... I think? I tried setting it to 16bit instead of 8, but only succeeded in making my ears ring when VLC played the outputted audio as full volume digital noise I know <@386612331288723469> was trying it too, storing audio as JXL

Didn't know that JXL can be used as an audio visualiser

not strictly ffmpeg, but zimtohrli seems easy enough to use: https://github.com/google/zimtohrli

https://news.ycombinator.com/item?id=40297280

What food is it related to?

Zimt is german for cinammon

Ohr is ear

I wonder how well the (non-ai) content-aware fill from photoshop and the like would work as a predictor

i guess pretty well? they'd just be super slow and require a lot more surrounding pixels than you get from the scanline pattern

It was around 1.25x the size of FLAC

So not terrible

Also not good though

If I remember it correctly

This was a while ago

Lossy worked fairly well, but even lossless there was still something wrong with the format conversion

Is that really for https://en.wikipedia.org/wiki/Palmier ?

I don't quite get your question. Zimt is indeed cinnamon and Ohr is ear, resulting in the word cinnamonear in english. This pastry does look like it could be called a cinnamon ear, but Zimtohr is not a name for a pastry in german, its a neologism. Palmiers are also called palmier or pig's ears here

I still want to eat a real zimtohrli

The goal is to eat all of the "-li" foods

then you'll have to visit switzerland xd

get yourself a zuckerli

https://github.com/dropbox/avrecode

has anyone used or tested this mid- to large scale?

It's the not actually used swissified word for the German word for these things:

Discord fucked up...

Those are called Zimt-Öhrchen in the region in Germany where I live (Rhineland), so...

proof: https://www.chefkoch.de/rezepte/2903841442748009/Zimt-Oehrchen.html

Not only have I never heard of that, googling for zimtohr also didn't yield any result lol

But good to know I guess

yeah, "zimtohr" won't get you anything

the -li in swiss german translates to -chen (standard German), or -lein (southern German), or -l (Bavarian/Austrian), or -le (Swabian), or -el (some Middle German dialects), or -erl (also Bavarian, Austrian), or -ke (some Northern German dialects)...

there are definitely even more of those

except jpegli

the xl from jpeg xl is also, while not too common, a sound used in some german dialects It would be pronounced jpegsel

It stands for Extra Larger

Ten L under Apple's pronunciation

And there's JPEG Ten R which isn't compatible with my iPhone "Ten R"

Nah then it would be "forty"

Since XL is roman numerals for 40

Maybe that's a cool name for the new format. jay peg forty.

Do I look like I know what a jay peg is? Buuaah!

Google Chrome Team: why do we need 40 if we've had 51?

Hmm

uhhhhh

A super high dose of estrogen is great for male vitality

~~take both at the same time~~

That's two packages of the same thing :p

LOL

someone has been naughty xD

SMV, the actual SMV, is kinda useless IMO, it has some of the right ingredients, and it does seem like someone with a brain made it, but the dossages are way to low to be usefu;

https://magazine.labdoor.com/labdoor-special-report-we-tested-the-infowars-supplements/

Just buy anabolic steroid powder on the black market and mix whatever you need yourself.

based

though HGH supplementation is actually legal and you can get it perscribed [av1_chad](https://cdn.discordapp.com/emojis/862625638238257183.webp?size=48&name=av1_chad)

I'll be ordering semi-legal estradiol enantate from Ukraine this year.

isn't that just birth control?

> Estradiol enantate is used in combination with the progestin DHPA as a once-monthly combined injectable contraceptive for women in Latin America and Hong Kong kinda

it's an ester of estradiol that the body slowly converts to regular estradiol (half-life in the body about 6-7 days). Estradiol is the most common form of estrogen in the body.

It can be used for anything any other common form of estrogen is used for. That includes birth control but also hormone replacement therapy.

😂

what?

Just that photo lol

I didn't realize there were a lot of messages below since then because discord doesn't scroll to the bottom

ah ok

Mixing them together, yes 😂

Combine their powers, become god

reminds me of this: https://youtu.be/C8dfiDeJeDU?t=9m1s

I hate it whenever people start talking passionately about politics... it's like sheep talking passionately about their favorite farmer

Regardless of your opinion, you're always wrong

~~the only good opinion is that the only good politician is a dead politician~~

The only winning move is truly not to play

https://fosstodon.org/@mmstick/113952008189644564

well this is interesting

there were some concerns with memory usage in cosmic, so that's good to see

just don't use glibc's malloc <a:tapshead:482628289941340200>

mimalloc seems like a really really good one

aaaa they are still using an outdated version of LuaJIT I was going to try and contact the devs to ask them to update it many months ago but I forgot

Noita is currently on LuaJIT 2.0.4 which lacks a lot of nice things like trace stitching of uncompilable functions

ancient forums are interesting

literally archeology

You say archeology, but this is from 2017 lol

which was over 7 years ago

> The use of other than the specified type could may cause fire or electric shock, which is extremely dangerous. I like how they try to convince you that an electric shock or an unintended fire are not good

Never know when you might want a tingle or some extra heat to get your day going

https://fxtwitter.com/khwampa/status/1889381078173098011

https://www.philippineairlines.com/ph/en/newsevent-listingpage/travel-advisory/cebu-catarman-and-more-cebu-siargao-flights-starting-march-1.html

note to self: don't ask ChatGPT to suggest relevant additional references that can be added to a paper. It will just hallucinate references that don't exist 🙂 Like this one: J. Sneyers, J. Alakuijala, "Butteraugli: Perceptual Image Similarity Metric," arXiv preprint arXiv:1609.08675, 2016.

sounds somewhat plausible but that paper does not exist 🙂

or this one: F. Bellard, “Better Portable Graphics: A Modern Alternative to JPEG,” ACM Multimedia Conference, 2014.

or this one: R. Xu and L. Liu, "Comparative Study of Hardware Decoding for Next-Gen Image Codecs: AVIF, JPEG XL, and WebP2," IEEE Transactions on Circuits and Systems for Video Technology, 2023.

If you search for "As a large language model" on google papers you will find thousands since ChatGPT started

(I think it was called google papers?)

google scholar?

Probably, yes

That led me to [this](https://dl.acm.org/doi/10.1145/3658664.3659648) paper, which seems... > We compressed the images (see below) to 10 different file-sizes between 5kB (compression ratio 94) and 100kB (compression ratio 4.7). > We clearly notice AVIF, HEIC, and BPG taking the lead across all considered compression ratios, while JPEG is clearly the worst. > WEBP is ranked fourth, while surprisingly JPEG XL is clearly inferior to WEBP. Concerning the relation of JPEG2000 and JPEG XR against JPEG XL we notice a different behaviour: JPEG2000 (for low bitrates) and two JPEG XR variants (for high bitrates) outperform JPEG XL. Strange...

"SSIM"

I think we have the answer right there. although I haven't actually opened the paper

this is better than the person that "wrote" a paper, a paragraph of which started with "As an AI language model," <:galaxybrain:821831336372338729>

yes this lmao

somewhat related: i'm working on a browser extensions that will allow users to vote whether web pages look AI generated, and when visiting one such page the extension will warn you in advance

for pages with no votes, will it use AI to extrapolate a score?

It will extrapolate based on reports for other pages in the same namespace that's controlled by the same author ```rust enum AuthorControl { // the user responsible for the content on this page does not control this domain or subdomain. // examples: // - medium.com/@johnsmith/content // - github.com/johnsmith/content // - docs.google.com/document/content PageControlled, // the user responsible for the content on this page controls this subdomain of the website. // examples: // - johnsmith.medium.com/content // - johnsmith.github.io/content SubdomainControlled, // the user responsible for the content on this page controls this entire website. // examples: // - johnsmith.com/content // - medium.com/about-us // - github.com/work-for-us WebsiteControlled, } ```

there's no AI used anywhere :p

Probably just have an icon show to say "Based on domain instead of page" or if there's not many votes for a page

yeah that's the plan, i was thinking of "we don't know about this specific page yet, but other pages from this author have this verdict. after reading this page, submit a verdict to improve the accuracy of these warnings"

i also have a fun idea for invisible rate limiting

in the off chance people start submitting spam verdicts, and scraping the database to filter AI content from the dataset used to train AI, there's a "shadow database" that just contains submissions that were identified as likely spam or inaccurate based on some heuristics. and when I detect an attempt at scraping, it'll start mixing in results from the shadow database to poison the data

so if AI companies were to attempt to use the API to filter AI content from their training data collection, and i'm able to detect that, then it will serve bad data from people abusing AI to write spam articles, crippling their product with a poison of their own making

usually i'm all for open data but in an attempt to clean up the mess from AI companies, i don't want them to be able to use it to purify their data and enhance their products which are responsible for the mess in the first place

if i get a *lot* of data then i won't have to rely on this part of the submissions, and instead i can dynamically cluster namespaces based on the types of generative AI used and the amount of pages that are AI generated compared to those that aren't

maybe even with URL path granularity

Clearly, this paper is not doing a great job at codec assessment 🙂

maybe it was written with ChatGPT

it seems like ChatGPT is the best at coming with paper titles though, so it can be used in that way 😁

my hypothetical metric shall be called PISM

It even is best at coming up with potential collaboration partners

I guess Zig ramblings aren't too on-topic so better here. This time, a parser for JPEG XL image header with the bundles' shapes defined using a schema, parsing code generated at comptime and it prints the name of each field and starting bit offset. Comment out the std.debug.print statement in Reader.zig and all bloat related to knowing field names at run-time just completely vanishes from the compiled binary. https://gist.github.com/jjrv/cb373bcbde3f5d6287977f5ad0c510f8 Goal is to compile to wasm and put it somewhere under github.io where you can drag a local .jxl to the page and see an analysis of file structure.

Looks like the purpose of Reader.zig (BundleReader would've been a better name) is similar to https://github.com/tirr-c/jxl-oxide/blob/main/crates/jxl-oxide-common/src/lib.rs It might serve to compare Zig comptime and Rust macros. I added lots of comments to the Zig code to clarify it.

Are rust macros very different?

Rust macros are macros, while Zig comptime is more like a really powerful constexpr.

Does Rust also support constexpr?

limited and gradually expanding support.

I like constexpr...

But people want a language that's clean and simple like C but more smooth and convenient with less footguns. C++ offers some convenient features but it inherits all of the problems of C and people also want a language to be more simple and coherent and easier to read someone else's code than C++

no one is really trying to make a more convenient less hazardous C

With the same essential simplicity and timelessness and universal utility

Zig does try to do just that

I hope so! It needs to be easily callable from C and convenient to verify that a C header file is compatible with a zig library, if it's to be successful. C is basically the polyglot tongue right now.

It's not safe like Rust but safer than C thanks to slices instead of pointers and not letting you accidentally ignore errors or null

I think safety features can be added incrementally to a language

It's kind of more timeless than C since it can target all supported platforms regardless of what you're running on

In all official binaries at least

I think one of the design directions of Zig is to remain simple and readable. That alone is very commendable. In order to be a viable alternative to C it needs to actually be more convenient and practical than C and that's a tall order

Even though C has a lot of practical flaws

A replacement needs to capitalize on fixing those flaws while keeping the simplicity and convenience

you can do that with Clang too

A little known fact about Zig is that it was created as a C++ replacement, not a C improvement.

Well, the Clang on my Ubuntu setup can't immediately compile Windows or OS X apps. In Zig it's difficult to make it not capable (in comparison)

Zig bundles libc headers for several platforms. That is useful, but it's not more than that.

And they made the choice to always bundle a linker that handles multiple things.

So you just download the tarball, decompress and a cross-compiling setup is all ready to go

They did made a macOS linker, but the rest is just regular LLD (they're making their own linker for fancy features, but it isn't needed for the cross compilations)

Zig, clang, and rustc can all compile their respective languages to any target no problem, but for any target where you need to use the target's libc, the headers and sometimes the linking can be annoying.

zig cc is a wrapper around clang that includes all the libc headers for several platforms, and some other smartness. technically speaking, it can be used for any language. E.g. you can use zig cc to cross compile Rust code.

https://www.phoronix.com/news/Karol-Herbst-Nouveau-No The Rust drama fallout continues. (And oh god, the comments are painful to read. 🙄)

LTO is pretty nice...

It shouldn't be optional

Nouveau has not had automatic reclocking for like a decade now.

I didn't know they even still develop it

I thought it was basically a dead project already. And while I agree talking about a "thin blue line" is absolute cringe, so is asking for an entire blog post complaining about it to be included in the kernel source tree

I don't think Zig's new linker does LTO. The point of it is to (re-)link executables as fast as possible

a tool like this would be really cool!

i'm hoping this brings some attention to Redox OS. imagine if people the likes of Hector started work on Apple Silicon GPU drivers in Rust for Redox

especially since the at the core this drama revolves around inclusion of Rust components

Redox OS having a Linux compat. layer in userspace and Direct Rendering Manager drivers on kernel-space would send Linux distributions crashing.

Although, adding DRM drivers to Haiku OS is more realistic IMHO.

i think Redox sort of has a Linux compatibility layer

they even changed the way file paths work on Redox to something that more closely resembles now they work on Linux to make that compatibility easier

i don't know what state it's in but it's certainly a goal to be able to run Linux programs on Redox without too much hassle

I believe redox is going for VM

huh, not from what I remember. i might be misremembering or maybe they changed course. i haven't been keeping up with its development for a while

I know they had more close to linux support for making it easier to port programs, but I thought that was the extent of it

ah that might have been what I confused for linux compatibility

i still really enjoy the fact that this is a thing at all

I still want a relibc linux target lol

(and this lol)

what would that do? relibc is a port of libc right?

hm does that mean it would use a rust version of libc

port is not the right word, it's an implementation, like glibc or musl

but yes, it would be an alternative, rust target to -glibc or -musl

i'm not too familiar with all this, are those libc implementations which are used internally by for example `println` and file IO etc?

Yes. The Rust stdlib uses libc to interact with the kernel.

right

How about a target based on https://github.com/sunfishcode/c-ward/tree/main/c-gull ?

Or even better, not using any libc at all on Linux and just using Rustix.

to lacking, they already have mustang and eyra to build apps using rustix and other stuff

Yeah but like Mustang was literally created as a new Linux target and no one seems to have cared much, so what does relibc get you?

Especially since relibc also lacks stuff.

relibc is a lot closer to actually being usable then mustang/eyra/c-ward is, that is all

I assume you can still use and compile C on redox... the C standard library is the interface between C and basic io and OS resources

Linux doesn't come with a C library interface for some reason. They actually left such basic and essential work for someone else to do because of abject slovenly laziness so you need to obtain the C library interface separately from an unaffiliated third party. Usually it's GNU glibc or the more modern musl but there are a few others too.

Same goes for other utterly basic and essential utilities such as filesystem creation/repair tools, /bin/sh and even /bin/init

Interesting article https://xav.io/blog/rust-formal-verification/

only just learned about this: https://tysontan.com/gallery/gallery-mascots/libbie-2017/

nice

less nice: its story, told below

What the heck

What is wrong with people

Personally it's a bit hard on my eyes, but sounds like Chrome and JXL :P

It's sad

Sometimes life experience makes us a bit jaded but we should try to preserve and protect our inner child and never fully lose our own innocence

me: “whew, I’m finally done with tidying up and cleaning, now I can relax and play some video games” the game:

https://tenor.com/view/club-penguin-clean-fast-gif-652211122523426052

did windows implement a new design for taskbar hover animations?

this feels different

For me it's flush with the top of the taskbar

It did.

Hi, they didn't invite us, just politicians making photos, I see ANS poster in top left here https://www.sejm.gov.pl/Sejm10.nsf/komunikat.xsp?documentId=501B8003F8AFE1ECC1258C34006F57C5&symbol=M_WYDARZENIA_KOMUNIKAT

https://www.facebook.com/photo/?fbid=1188563146602976&set=pcb.1188564633269494

Too bad. As usual, the politicians get to take the credit while they're usually the last ones who actually deserve it 🙂

Indeed, at least JPEG XL is also advertised there: "JPEG XL – a new image file format, which might supersede JPEG as the basic image and photograph storage format after 30 years of use, since it enables reducing the file size to, since it enables reducing the file size to a third of the original without perceptible loss of quality. This format has already been used by Apple and Samsung."

Politicians patting themselves on the back and gloating over the achievements of others, water is wet and the pope loves kids.

That's a pretty good summary of jxl...

the irony

o h

no idea if the person who created this meme intentionally exported to jpg, but if they did then [galaxy_brain](https://cdn.discordapp.com/emojis/905916982145220658.webp?size=128&name=galaxy_brain)

A brand-new product that supports decoding JXL https://www.apple.com/iphone-16e/

> Built for Apple Intelligence. this being the second sentence on that page is hilarious

does this replace the iPhone SE?

Yes , although it is a good bit more expensive

https://www.threads.net/@wongmjane/post/DGQwBjjPgYa

That's a revolutionary upgrade

Worth every penny🤑

i fixed it

Question: Isn't the Apple Polishing Cloth a revolutionary product? ChatGPT: > Haha, if by "revolutionary" you mean a $19 microfiber cloth with an Apple logo on it, then sure! 😆 It definitely sparked a lot of conversation (and memes) when it was released. > > To be fair, Apple does claim it’s made of a "soft, nonabrasive material" specially designed for their displays. But at the end of the day, it’s still just a cleaning cloth—one that costs way more than most alternatives. > > Do you think it's worth it, or is it just Apple being Apple?

I think it’s helpful if you have a nano texture display, stupid otherwise

I'm so glad they released a new one that's compatible, now I can finally upgrade

I hate having to rely on third party aftermarket gear

Apple branded products always work best

https://www.youtube.com/watch?v=8nawfm3uVEk

https://www.loudersound.com/features/tangerine-dream-phaedra > A particular low point occurred when Froese witnessed the Manor’s resident Irish wolfhound amble in, raise its leg, and urinate against the band’s Mellotron.

ouch

well, I did end up having some unpleasant dreams, albeit not about that tone-mapped image

in one of them, Hagrid (I think?) cast magic spells to assassinate a bunch of Hogwarts professors

for one of them, by evaporating all of his body water

in another dream, the venom symbiote from Spider-Man created a few million offspring who went on to bond with correspondingly many people, and had them amass in a giant pile

||welp, just resumed my game of Marvel’s Spider-Man 2 and that’s pretty much what happens next – seems my dream was premonitory||

Gotta love it when your phone with a 16320x12240 sensor shoots 4000x3000 in 12MP mode and not 4080x3060 <:KekDog:884736660376535040>

Depends on the application (sometimes)

Try OpenCamera

OpenCamera does have a proper scaling option, but doesn't really work properly for me

It can't even record in 4K60 and custom shutter speed is limited to the point it's often useless. And somehow the output is also often disappointing

maybe try motioncam? there's a free demo version

True, but the image output is actually correct, resultion wise lol

I just use 50MP mode in my default camera app now which does use correct scaling <:KekDog:884736660376535040>

Motioncam also saves "correctly", yes

My phone removed my resolution options in a firmware update 💀

lmao

I have the option of JPEG, JPEG-Large or RAW

absolute peak android rom moment

I am a 4032x3024 / 8064x6048 fan personally

reverse kaizen

my old phone did something similar, it removed the possibility to turn off digital video stabilization

can this output actual 200 MP photos?

that's the new samsung, isn't it? i guess i'll look into that more... i really can't imagine 200 MP being that much of a game changer on a tiny sensor with tiny optics. if they're actually good, that makes it all the more impressive

Mine just outright removed it, my only guess is that 720p is stabilized digitally, since it only allows 30fps too but 1080p runs at 60fps

It's kinda useless in most cases, but can be nice when there's a lot of light I guess

It's pretty useless with my indoor lighting

it looks really bad

The scientists were so obsessed with if they could, they never wondered if they should

It's fun to have the option available I guess

i'm curious, can you hold your like an index finger's length away from the screen and take a really still picture? i wonder how it's going to deal with the display pixels

The quality can be slightly better than at 50MP, but only when there's a lot of light and you can hold the phone still

when you zoom in it does look like there's some detail being added back in the background which is supposed to be blurry, which is pretty funny

Also no RAW capture possible with 200MP

Just JPEG

Yeah, RAW only goes up to 50MP for some reason...

oh wow!

Funny that I can see the subpixels

It's a 4k 32 inch monitor

That's because it would be a 5 GB DNG, assuming 10bit

Worth it™️

I have 1TB anyways <:KekDog:884736660376535040>

😔

Would it not be 500MB? A 50MP DNG is around 120MB

that's glitch art of my comfort character

I was thinking uncompressed, and napkin maths

200MP is not worth it in any regard imo, 50MP also don't look great, with direkt sunlight even

Filesize sure jumped up

I should try making a 'proper' JXL Bayer image again at some point...

200MP can be worth it tbh, but only in some very rare cases. And it has to be resized for sharing, so for that it's kinda shitty

I have tried tinydng with non-linear DNGs before, but could not verify if it was proper, since no software on linux supports JXL DNGs yet. Well, at least the filesize was massively reduced

it's doing a lot of processing for sure, phone cameras do that regardless, even when not shooting super high resolution

Wish I could shoot lossless DNG at 200MP and then stack them. At least ISO does go down to 12 now

i got so used to pictures from my digital camera that when i took a picture with my phone i was confronted with water coloring ass texture retention in the sky in the corners and a WAY overdone attempt at cramming all that dynamic range into an SRD image while messing with the contrast in places

I don't mean DNG, I mean kCFA that Adobe didn't use

as far as i'm aware phones that output DNG at all still do a bunch of processing, like combining multiple shots into a single image

so it's not really 'raw' regardless, but it is uncompressed. at that point i'd rather they just give me an option to shoot in png or some other lossless format

Not Samsung I think, otherweise their raws would look better lol

Pixels and iPhones do though

Though, I know non-linear works, because this is the JXL file inside. Lossless is actually 25% smaller than lossy too

Think that's the difference between their stock camera DNG and Expert RAW DNG

Yeah, it looks funky indeed. Sadly I cannot edit them, since no Photoshop or something similar

Expert Raw uses a JXL payload

But the app itself isn't great

oh huh! i should check if my Fairphone 5 (which doesn't *officially* support outputting raws) does it then. maybe i can test it by getting a very fast moving object in frame and see if there's ghosting

i've never seen ghosting in general actually, i wonder whether if it does it in the first place

I did the debayering myself in Krita by separating and aligning the planes. Gives a rough idea of what a kCFA JXL would look like natively while still being a RAW file

Nice!

But probably too much effort for normal editing

Especially for larger amounts of photos

Could possibly be automated though 🤔

My phone is probably too old to process RAW files, though that's my only option to avoid smoothing and sharpening now anyway. Takes uncompressed DNG with no previews

Which camera app are you using for uncompressed DNGs? OpenCamera?

That was purely a proof of concept for the 'native' JXL RAW files. Since it's really just the 4 bayer planes and a boatload of metadata, which JXL can handle by itself

My normal camera app. The CPU is too slow to compress them :P

Heh, 200MP DNG works now apparently