JPEG XL

"yeah I need to take these with me to test taking shots of the aurora on the other side of the country be back in a week or two"

Maybe if you give them your firstborn as a guarantuee that you'll come back lmao

I'd assume in Switzerland they would be quite willing, mostly because if you never give it back then you have a big problem xD

incidentally, I think you’ll have no trouble recognising where the test shots from that day were taken

but the signal to noise ratio should get better when downsampling, right?

Speaking of noise, it was just now I realised that JXL ISO noise influences the encoding choices and changes filesize. I always thought it was just an overlay

just asking since, if higher resolution sensors aren't (much) noisier than lower resolution sensors on a per area basis, then I would expect downsampling to make the image less noisy.

There are many ways to do downsampling. Some are smoother, some are sharper, most reduce noise but there are some that attempt to preserve detail including noise.

Just doing averaging reduces noise but it's also too smoothing to my taste.

averaging would be like if you had a different sensor of the same size with half the vertical and horizontal resolution, right?

I've thought about reducing 24 mp shots to 6 MP shots that way if they're not going to be displayed at 100% scale anyway, but image viewers might do downsampling like that themselves so I'm not sure if that'd be worth it. It seems better to take a burst of shots and combine them to reduce noise, for still subjects at least

basically, the signal adds up linearly but the noise adds in quadrature (sqrt(a² + b²)) so only grows with the square root of the number of pixels, and therefore the SNR grows with the square root of the downsampling factor

so if you downsample from 60MP to 15MP, the “new big pixels” have roughly twice the SNR of the “old small ones” (4 times the signal and twice the noise)

I have the impression that you might be counting the downsampling twice, though

this effect of downsampling is more or less why we can say that they aren’t much noisier

(or at least correlates with it in some sense)

there is the effect of denoising, though

at efforts ≤7, it should be, but I wouldn’t be surprised if it interfered with the butteraugli loop of efforts 8+

Oh yeah, of course. Thanks :)

This was effort 4 and 7

oh, surprising then

I would have expected the ~10 bytes of noise LUT to be the only difference

Higher ISO lower filesize, so it seemed to be discarding the pixels somehow

I don't *think* I am? Just did a double take when you said that the noise increases. Then I realized, oh yeah, but the signal presumably goes up faster. Just needed confirmation cause I'm still rather unsure about this topic.

Ooo, is this denoising the image to attempt to "replace" the noise with synthesized noise, or is that not implemented yet

ah, right, I misinterpreted what you meant

you meant that as “that must logically mean that downsampling reduces noise”

I thought you might have meant “that would mean that with downsampling, they would be less noisy”

Yeah, former.

I wonder if some of the a7S III's performance is simply due to superior electronics compared to the other cameras, rather than the differences in resolution.

I’m not fully convinced that the performance of the S series is all it’s cracked up to be in the first place

also, for what it’s worth, in the α7S III, it’s apparently a quad bayer 48MP sensor

https://www.sonyalpharumors.com/surprise-the-sony-a7siii-actually-has-a-48-megapixel-quad-bayer-sensor/

😮

I'm reading up on dual pixel AF and [this article from Canon](<https://www.usa.canon.com/learning/training-articles/training-articles-list/intro-to-dual-pixel-autofocus-dpaf>) says that the two diodes that make up one photocell also have their signals combined during capturing. I wonder if that gives you some free higher SNR as well, at least for area of the sensor that has DPAF pixels

I'm now *also* wondering if DPAF means you get a free depth map which would be nice for future cameras that natively shoot JPEG XL, but I'm not really sure how signals from these pixels are processed

Actually, I think the Google Pixel did something like this? I remember seeing a blog about building depth maps from dual pixel autofocus

[yes, it does](https://www.canon-europe.com/cameras/eos-r5/dual-pixel-raw-mode/#:~:text=By%20comparing%20the%20very%20slightly%20different%20views%20seen%20in%20each%20of%20these%20images%2C%20a%20depth%20map%20can%20be%20embedded%20in%20the%20RAW%20file%2C%20meaning%20subtle%20adjustments%20can%20be%20made%20to%20the%20end%20result%20based%20on%20this%203D%20data%2E)

Okay I found it, it really is just stereoscopic solving to get a depth map from DPAF pixels

the whole sensor is DPAF pixels, but each dual-pixel is two smaller pixels, so no free SNR

I don't understand, doesn't more samples per pixel mean lower noise? You said before that merging 4 pixels into 1 does reduce noise compared to if you had just 1 pixel, no?

It’s like averaging multiple samples in science

compared to only one pixel like the 4 you are merging

not compared to one big pixel of the same size as the 4

A lot of people online say that increasing your ISO only improves the SNR if a camera is iso-invariant *or* dual-gain, and never otherwise. I'm going crazy.

Ahh right, that makes sense

one big pixel that receives 100 electrons on average, with 1 electron of read noise: 10 electrons of shot noise + 1 electron of read noise = sqrt(101) = 10.05 electrons of noise, SNR = 100 / 10.05 = 9.95 = 19.95 dB four smaller pixels that each receive 25 electrons on average, also with 1 electron of read noise: before downsampling: 5 electrons of shot noise + 1 electron of read noise = sqrt(26) = 5.1 electrons of noise, SNR = 25 / 5.1 = 4.9 = 13.8 dB after downsampling: sqrt(104) = 10.2 electrons of noise, SNR = 100 / 10.2 = 9.8 = 19.8 dB

that's a very small difference

it gets a bit larger at very low signal levels where read noise makes more of a difference (but it might be beyond the point you would consider usable anyway)

what DxOMark calls “Print” SNR/DR is after downsampling to 8 MP

“Screen” is at native pixel level

most reviews that claim an advantage for lower-resolution sensors look at 100%, as in the bottom image

controlling for this paints a different picture

what's "measured" ISO?

to be fair, the α7S III does do slightly worse than the II, which some people have taken as evidence that downsampling a higher-resolution sensor is not as good, but the performance of the α7R V argues against that interpretation

a useful metric with a terribly misleading name: pretty much 78 lx·s ÷ _H_ where _H_ is the focal plane exposure at which the raw data saturates

which means that looking at the SNR at 18% of that lets us compare cameras at the same absolute light level, which accounts for potential differences in light collection efficiency (“quantum efficiency” or QE)

(historically, one possible method for computing an “ISO speed” for a digital image was 78 lx·s divided by the focal plane exposure that would saturate the output _processed image_, not raw; DxO took that formula and applied it to raw, which has its uses but has essentially nothing to do with ISO)

what is lx and s?

lx is lux, s is second

ah

thanks!

actually, what exactly does focal plane exposure refer to?

https://en.wikipedia.org/wiki/Exposure_(photography) > In photography, exposure is the amount of light per unit area reaching a frame of photographic film or the surface of an electronic image sensor. It is determined by shutter speed, lens F-number, and scene luminance. Exposure is measured in units of lux-seconds (symbol lx ⋅ s), and can be computed from exposure value (EV) and scene luminance in a specified region.

approximate formula:

righto, thanks. The "focal plane" qualifier threw me off.

when Luca and I were trying to determine what sort of HDR tone mapping his TV performed, we used this formula in reverse to estimate the luminance with my camera 😁

nice :)

Gonna pester you with one more question: which graph on photonstophotos is the right one to look at for determining whether a camera is iso invariant, and how should the graph look? I've seen people online in discussions link to all three of dynamic range shadow improvement, read noise in DNs, and input-referred read noise.

input-referred read noise should look flat

“dynamic range shadow improvement” being close to 0 might be an acceptable proxy

but the dynamic range it refers to is the “photographic dynamic range” metric that Bill Claff came up with, which uses an arbitrary SNR threshold as the lower bound instead of read noise

huh, which cameras are even ISO invariant then? I've yet to see one that people say is ISO invariant that is flat on that graph. And people talk about each of the gain ranges in Sony cameras as being ISO invariant to itself. So supposedly setting the a7RV to 320 ISO and brightening in post should look the same as as shooting at a higher ISO. I've seen a few people demonstrate this but they weren't super rigorous.

_Physics of Digital Photography_: > Although figure 5.21 shows that higher ISO gains yield a higher SNR in the low exposure regions of the image, the advantage gradually lessens each time the ISO gain is raised. Eventually a value is reached where the upstream read noise dominates and a higher ISO gain would bring no further advantage. For the Olympus® E-M1 used to produce figure 5.21, all curves above ISO 800 lie almost on top of each other. The precise ISO setting _S_ above which the corresponding ISO gain would bring no SNR advantage compared with applying digital gain in postprocessing is referred to as the _ISO-less_ setting. > […] > Some recent cameras have such low levels of downstream read noise that the ISO-less setting is very low and the camera can be described as being _ISO invariant_ [35].

ooh, I should get a copy of that

using the terminology in that excerpt, the people you mention might have meant that ISO 320 is the α7R V’s ISO-less setting

(but ultimately, looking at the input-referred read noise curve gives you more information than any categorisation into “ISO-invariant” / “not ISO-invariant”)

both ranges are closeish to flat

(d5000 for reference 😄)

what they usually say is 100 is the start of one iso invariant range, and that 320 is the start of another iso invariant range. So presumably they mean that you should shoot at 100 unless it would've been too dark for 300, in which case go to 320 and stay there.

yeah, given that curve, that’s probably how I would use it too

100 or 320 and nothing else

I'll agree with your statement that this could probably be talked about it better ways than saying that things either are or aren't iso invariant.

I don't think your rule of thumb regarding the shadow improvement chart works though

I think this reflects that those cameras aren’t really “ISO-invariant” in the strict sense (the sense where the ISO-less setting is the base setting)

ah

someone I admired would call such binarisation “mutilating barbarism” and “Procrustean deviancy” 😁 https://x.com/JamesMaloneLee3/status/1317756706156400640 https://x.com/JamesMaloneLee3/status/1335629601456926725 https://x.com/JamesMaloneLee3/status/1409056750846844928

hooray

ok I made a slight mistake

I forgot to look up when the C/2024 A3 comet was going to set in my location...

So if I'm gonna head out in the opposite direction of the airport with my camera, I'll have to aim for something else!

reckon I can get a magnitude 13.1 comet with my camera? lol

Guess I'll just have to try tomorrow

the a7S III looks kinda funny in the input-referred chart

the plastic bit on my tripod holding the mounting part to the legs just cracked as I picked it up. That's kind of annoying. But at least it didn't happen when the camera was on it.

sheesh

maybe I can glue it back together...

Cool, what ISO sensitivity was this taken with?

originally 12800 but brought back down 2-3 stops

I went and did some evening bird photography. The sparrows were beyond my ability to track, but I got some nice shots of magpies.

<:BlobYay:806132268186861619>

I think I should buy one of those 18% gray cards so I can better adjust the color balance

I bought https://www.greywhitebalancecolourcard.co.uk/

ty for the suggestion

finally pulled the trigger on Canon’s tilt-shift 24mm

I’ve been eyeing it for close to a year

👀

part of why I waited for so long was that I wasn’t 100% sure whether the 17mm, the 24mm or the 50mm would be the most suitable for me

(I’m still not 100% sure, but somewhat closer)

Are there any reviewers or sites other than DxoMark that try to assign scores for sharpness?

Lens rentals sometimes has a look at lenses

Mostly Blogposts

Not a lens review but a nice article nonetheless

https://www.lensrentals.com/blog/2017/07/experiments-for-ultra-high-resolution-camera-sensors/

It is a bit older though

Wait, they have a section dedicated to rewiews now https://wordpress.lensrentals.com/blog/category/reviews/

thanks

i went down the lens sharpness hole and my conclusion was that everything is fine

I'm trying to figure out which lenses are actually sharp enough to make full use of a 61MP sensor, especially if cropping into a corner

hm, aight

Lensrentals did some comparisons with older and newer lenses and concluded indeed that old lenses are perfectly fine

Sony claims that their GM lenses are good with 100MP, but this is very possibly just marketing

because even the sharpest lens is bottlenecked by everything else

one-dimensional scores, I’m not sure, but various sharpness numbers can be found on lenstip or OpticalLimits

more detail than just one number is fine, indeed better for me, thanks!

True, also most camera still have that filter in front of the sensors that kind "smoothens" the ingoing light, what was its name again?

incidentally, see what Roger Cicala thinks of DxO’s “perceptual megapixels” https://www.lensrentals.com/blog/2019/10/more-ultra-high-resolution-mtf-experiments/#:~:text=Appendix%3A%20Why%20Perceptual%20Megapixels%C2%A0are%20Stupid

I tried the sony gm 2470 and photos ended up very similar to the 2070 G, ultrawide G and even the kit lens

If you are looking for sharpness above all then macro lenses are among the greatest you will find

otherwise fixed lenses

yeah for landscape youre bottlenecked by the air

for zoom lenses the wider end is almost always the sharpest

100% true

that's how i ended up on getting the lightest thing possible

because everything else doesnt matter

Gotta get that Sigma 500m 2.8f https://www.nikonpassion.com/wp-content/uploads/2012/10/sigma_200-500_01.jpg

I'm gonna put that $100 Viltrox 28mm pancake on a $3000 camera 😵‍💫

lightest thing with the fastest AF

Which camera?

and for sony E, only the latest sony lenses have the better AF

a7CR

The older lenses may be slower, but fast nonetheless

except for the 50mm 1.8f smh

a.k.a. the “Bigma”

Christopher Frost reviewed it: https://youtu.be/C9UnFxwww3Y

Yesterday was the first time I've managed to track a magpie at close enough distance to get a nice shot. I failed around 2/3rds of my attemtps. But when I succeeded, the autofocus failed to be fast enough around half the time.

I wielded that Sigma a few times, it is truly massive

With the pancake lens you mentioned?

Or a tele?

What I mentioned is what I want to buy in the future (though maybe I'll have different plans by the time I have the money)

So with my Nikon D5000 with a 70 to 200mm tele

some warning the big MP images are a pain to work with

you can ONLY use sony's software or latest lightroom

oof

the files are HUGE

you need at least UHS-II sd card

Good thing I'll be able to afford a macbook at the same time

you mean the pixel-shift images?

the regular images

native shots should be fine, right?

ah

Darktable and RawTherapee don't work with them?

they dont support ARW 4.0

Darn

I hate proprietary formats

And the sony software is crappy enough as is

you will ONLY use JPEG and TIF

bcause other formats are too laggy

DxO PhotoLab says it supports α7C R files except “RAW M and S”

honestly I quite like the JPEGs I get out of my 15 year old Nikon and presumably recent cameras are even better at producing JPEGs

yeah the jpegs are good

for this I mean when you are processing on your computer

lightroom takes AGES to load the images in the import dialog

Though I read that Sonys don't have any function to do raw->jpeg conversion on the camera, so presumably I won't be able to take a photo and then decide it needed to be stepped up or down a bit.

Lightroom is slow in general

for shooting you do jpeg+raw

yeah

Am I the only one who exclusively shoots RAW? xd

Camera JPEGs are a waste of space imo

made me think yeah i probably can not use jpeg

since the raws have previews

and load as fast as the jpeg

They should add a mode that does the same processing as the built in JPEG conversion, but stores it as a 16-bit JXL file so all the detail remains.

if you put those images in jxl your computer will explode

JXL support in 5 years maybe, sad

with streaming decoding it will be fine

so that's why I think getting the a7cii may have been a better experience

Ok, at least darktable was able to open an ARW for the 7CR

maybe I would have had the possibility of editing on the go

Do you ever shoot in the downsampled modes?

no

Anything like downsampling can be done easily in post

Greater control and better results

Sure, but if you want more manageable file sizes and don't care about losing the higher resolution, it works.

i paid full price for this so I may aswell use it all

Agreed

no blessing only curse

"Storage is cheap" I have almost a TB in my camera

yeah i carry 2 512gb cards

heck yeah

it's not expensive

I paid 25€ for one 512gb card

my early bought 64gb uhs-ii was 120€ lol

but well, early adopters always pay a premium

to be fair, DxO doesn't assign a perceptual megapixel score to just a lens, but to a lens and body together. For example, Sony FE 28mm F2 on an A7R II gets 35 P-Mpix, and on an A7R IV it gets 47 P-Mpix.

Lensrental uses machines that do not rely on cameras at all, ony the lens itself is tested

I forgot the name, it's been a while

using Sony’s software, you might

bench?

https://www.lensrentals.com/blog/2014/05/introducing-the-optical-bench/

with all that said: https://www.dpreview.com/opinion/4042117089/roger-cicala-why-i-dont-use-an-mtf-bench-to-test-my-own-lenses

Indeed, which sounds a lot more annoying than doing in on camera, considering how janky vendor software often is (and everyone seems to complain about Sony's camera related software). Doubly annoying because I'm on Linux 😄

My own camera isn't even supported by sony's mobile apps anymore

Very fun, i was on the road and wanted to remote control my camera, only to find out that the app had updated and quietly removed support for my camera

Oh and I doubly hate supposed features that require a vendor app to use. Like ok I can control the camera with the Sony camera whatever app, but what happens when the camera is 10, 15 years old? It'll still be very good, but oops now it's worse because of lack of app support.

Doesn't need to be 10 years, my camera is 6 years old and support was removed when it was 5 years old

Sony is also not very good regarding firmware updates, my camera received its last one three years into its life

Oof

Which in my case, with the a7III, means that I did not receive the update that removed the 29:59 limit for video recording for example

Not a large issue, but annoying nonetheless

my EOS R6 still receives firmware updates _but_ that still doesn’t include lifting the 29:59 restriction AFAIK

Man...

my catalog here <https://tanbou.grass.moe/2024/05/mygo/> is all taken with the a7cR

nice photos 🙂

thx, was most fun lining them up

some more phone crashers

from last week

nice

the New Yorker photo and the playground+tower photo might be nice to develop in HDR, they look like they would make interesting test images 🙂

Does developing HDR only work in Lightroom etc? What does the workflow look like? Or is it enough to export to rec2020?

there ought to be a way for lightroom to use a phone as preview

i can't get the hdr output in lightroom to work

https://www.filmpixmedia.com/laowa-55mm-and-100mm-f-2-8-ffii-tilt-shift-macro-lenses-finally-announced/

just today 😂

mfw no unlimited budget to satisfy GAS

Gotta buy them all™️

a secondary problem is having enough space to store it all

ooh

here's a photo I took a few years ago on my phone at 3:30 in the morning

Those look incredible

I love fog

^_^

I love anti-crepuscular rays

Here is a photo from last winter. Rare light pillars caused by ice crystals perfectly suspended in the air.

When I went to the grocery store, there were miniature light pillars literally less than a meter from my face

That's interesting! I never thought about it but it makes sense that the same light rays also appear to converge opposite of the sun

it's cool

Anyway completely different topic. So the 48 MP sensor in the a7S III has a quad bayer pattern like this, and is downsampled during readout as shown in the picture. This allows for faster readout rates, useful for video.

If you were making a sensor primary for still photography, but you wanted faster readout rates, would a downsample like this make sense? You get some funky overlap in the locations of the photosites of each combined readout pixel, which might lead to something weird? Probably fine though. And I'm sure you could find a better pattern for the green pixels.

I think the effective resolution would be lower this way, but that's probably fine.

in principle, I think it should be possible to simulate the kind of artifacts we would get

high-resolution input (photographic or rendered) -> simulate sensor with that pattern -> downsample as you describe -> demosaicise -> output

oh, that'd be a fun little project

from what I recall, one potential drawback of this approach is that with very small pixels, CFA cross-talk can be an issue

(light goes through one type of filter but ends up captured by a pixel that’s supposed to have a different filter)

ah: https://www.dpreview.com/forums/post/59967266 > Sony essentially eliminated the problem in the a7RII with the combination of BSI and low-profile microlenses, without losing the high fill factor.

Is chromatic aberration not something you can correct for if you know the properties of the lens? I noticed chromatic aberration correction in Darktable isn't part of the lens correction module, and it's something I need to hand-tune, it doesn't calculate the parameters for my specific camera + lens it seems

Oh also I updated Darktable and my lens is actually in the database now, which is awesome

iirc, there’s both

lensfun (used by the lens correction module) definitely has CA data (I’ve contributed some), and the module can apply them

but there might be an _additional_ module that can also correct them from manually tuned parameters

https://github.com/lensfun/lensfun/commit/5c78d634364abfe2b6d895a4ea3eb0be7367a294

more generally: https://lensfun.github.io/lenslist/

Oh, TCA is chromatic aberration correction? I tried the TCA option on its own in the Lens correction module and it didn't seem to change anything. Maybe my lens doesn't have TCA data associated with it

I just checked, my lens (Canon RF-S 18-45mm F4.5-6.3 IS STM) only has distortion and vignetting information associated with it, so that explains it

can darktable do hdr?

It can export in HDR yeah. I don't know if it has HDR in the editing view. I don't have a HDR display so I can't test

TCA is transverse (a.k.a. lateral) chromatic aberration which is the correctable kind (as opposed to axial/longitudinal chromatic aberration)

(as Wikipedia points out, “LCA” could refer to either lateral or longitudinal CA so it’s ambiguous and should be avoided)

https://commons.wikimedia.org/wiki/File:Comparison_axial_lateral_chromatic_aberration.svg 1: no CA 2: axial/longitudinal CA (different wavelengths focus at different distances) 3: transverse/lateral CA (different wavelengths focus on the same plane but at different locations on it; only occurs off-axis)

Ah, good to know

it can technically export to HDR, but without being able to see what you output before you export it, it’s hardly usable in practice

oh, that’s quite a rare combination

does darktable have a auto color button light in lightroom

usually, it’s distortion+CA but no vignetting

A bunch of modules have auto options. I think by default it goes derives the options it uses from the metadata in the raw file. It's free software and they have a flatpak, so it's pretty easy to try out if you wanna mess with it, but I'll answer that specific question in a bit

waiting minutes for avif export...

lightroom's avif is only seconds is this e0 or something

phone fails to load this 🤦

Ah well I'm very glad it has vignetting correction as it's *very* visible at 18 mm. I'm more okay with not having CA correction, I didn't even notice it until someone zoomed in and pointed it out to me, and the defaults for the manual CA correction module take care of it well it seems. Although it does also remove some of the vibrancy and color on things exhibiting subsurface scattering which is a bit of a shame. I'll probably leave it off or learn to apply it selectively

How do you go about collecting metrics like these for Lensfun? Do you have like a test pattern that you point your camera at and take many different pictures at different configurations, and the software using the data interpolates the rest?

probably the checkerboard

How does the camera do it? Does it have an internal dataset or is that data stored digitally in the lens?

my photos have the lens data embedded

the lens communicates the data to the camera

the lens has a firmware

Right, but how does the camera know which corrections to apply based on that data. I can also see data like aperture, focal distance, etc, but without the Lensfun database Darktable still doesn't know how to correct for it, so presumably either the lens communicates the rules for applying corrections to the camera or the camera has a dataset of these rules

https://docs.opencv.org/4.x/dc/dbb/tutorial_py_calibration.html

this is for distortion

but it's a single formula and just 5 numbers

but there's just... more data i guess

yup there is just more data

> Vignetting Corr Params : 16 0 64 256 576 1024 1568 2176 2880 3648 4416 5248 6048 7040 8288 9952 11488 > Chromatic Aberration Corr Params: 32 640 896 896 1024 1024 1024 896 896 896 896 896 768 640 512 384 128 -256 -384 -512 -640 -640 -640 -512 -512 -512 -384 -256 -128 -128 0 256 512 > Distortion Corr Params : 16 0 -10 -40 -89 -157 -243 -347 -465 -597 -741 -896 -1059 -1229 -1405 -1585 -1767 etc

I think I followed https://pixls.us/articles/create-lens-calibration-data-for-lensfun/

the raw files should contain the data in question, but darktable currently doesn’t read it

also, in proprietary raw formats, it might not be clear how to read it in the first place – but Adobe DNG Converter then makes it accessible as OpcodeList3 https://helpx.adobe.com/content/dam/help/en/photoshop/pdf/DNG_Spec_1_7_1_0.pdf#page=105

> WarpRectilinear: This opcode applies a warp to an image and can be used to correct geometric distortion and lateral (transverse) chromatic aberration for rectilinear lenses. The warp function supports both radial and tangential distortion correction.

Ahh

ah, right, I remember that module now – iirc, it kind of “cheats” in that it “corrects” TCA by just desaturating edges

instead of actually rescaling the colour channels so that they align again

Yeah that checks out with what I've seen

I find it slightly funny that several a7R V and a7CR reviews rave about how good the noise performance becomes when they put it into downsample mode

paraphrased representation of reviews: *zooms both images to 100%* "As you can see, the a7CR is noisier than the a7C II, but if we bring up the a7CR medium downsample, it looks really clean"

_effectively zooms the α7CR image 1.35× as much_ “as you can see, it’s noisier” `surprised_pikachu.png`

it's crazy too, they're open right next to each other, you can see how much bigger one is!

In this video, the a7S III looks very good compared to the a7R V, even though he shows the images at the same proper scale https://www.youtube.com/watch?v=HdPAfAMPnls very likely that there is some confounding factor I am missing that would explain it, though. The fact that the second gain stage kicks in at a much higher ISO is interesting, but I'm not sure how much it might explain.

because megapixels aren't everything!!!

I always remember when my artist friend from elementary school always says 2 megapixels is enough

a7S III has a 48 MP sensor tho!

it’s a bit hard to tell because they don’t look developed with the same amount of contrast

higher contrast would exaggerate noise

if using e.g. unsharp masking, one would need to make sure to adjust the radius accordingly

not use a radius of the same number of pixels

Ah, good point

I've waited a couple of days for fog, today I was able to try out my camera in this weather for the first time 👀

The previews mess up some of the details, so if you click "open in browser" you can make out some shapes in the fog

Niiice

as I was hoping, it seems that I can use the shift feature of the 24mm not only to correct for the keystone effect, but also, on the contrary, to exaggerate it

and thereby, get sort of an UWA effect without actually taking an UWA picture

hooray

Do you guys usually shoot with noise reduction with high ISO?

I have it on default, but a colleague said they always shoot with noise reduction off. I'm thinking they probably have a much better sensor since their camera is like three times more expensive

I know there's no one right answer, I'm just curious what you guys do

if you are referring to what Sony calls [“Long Exposure NR”](https://helpguide.sony.net/ilc/1720/v1/en/contents/TP0001629724.html), it refers to the camera automatically taking a dark frame of the same duration to use for subtraction

“High ISO NR”, on the other hand, would not affect raws

I shoot raw so I apply however much noise reduction I need for each image

(and haven’t felt too much need to deal with dark frames so far)

https://photographylife.com/long-exposure-noise-reduction#what-is-long-exposure-noise-reduction

I'm using a Canon EOS R100. I just mean the noise reduction it does "by default" which they call high ISO noise reduction. Long exposure noise reduction is a separate option that happens for exposures longer than 1 s, and it's off by default

Yeah that's a good point, I shoot JPEG + RAW so I can always create one without noise reduction if I want I suppose

I didn't know long exposure noise reduction worked like that, that's clever. I should have enabled it when trying to take pictures of the night sky. I assumed it would just do a normal denoise pass so I left it off

you can also take the dark frames yourself, which gives you more control (although it also means you need to use software that’s capable of utilising them)

it lets you _stack_ dark frames, for example

I'm going to try long exposure NR with the lens cap on. It should take two identical dark pictures that are subtracted from each other and then I should be left with only the random noise and no static sensor artifacts. If you then compare that to the same picture without long exposure NR, I wonder if that gives you some idea of the quality of your sensor

Interestingly the ones taken with long exposure NR seem overall brighter and more noisy compared to those without

The corners seem brighter, I thought it might be light leaking in past the lens cap but it's also possible it's the vingetting correction

First time I see the Atomium at night Didn't know there were lights on it <:KekDog:805390049033191445>

neat

but as expected from a phone camera, really high noise

What model of phone?

Xiaomi Mi Note 10

and that pic was taken with the 2X zoom camera

never buying Xiaomi ever again

too many SW problems

Huh I was expecting a Google Pixel because the filename starts with PXL

Did you use gcam?

yeah (a custom one tho) LineAgeOS, signed with Pixel keys <:KekDog:805390049033191445> So I have unlimited Google Photo storage lol

I thought they scraped that

mmmh, did not check for quite some time true at least it was the case when I started using it

first outing with the tilt-shift 24mm

(just tilt here)

I wonder if people would misclassify images taken with this lens as being AI generated because "the focus doesn't make sense"

very cool

out city 🌌

If you ignore the slightly clickbait title, a neat little video of a DIY camera with a single photodiode and a color wheel https://youtu.be/Y-WEyZHIDFo

theoretically you can replace the little camera with an actual camera and use it to get automatic panoramas?

I'm quite confused by something in this article, <https://clarkvision.com/articles/iso/>, specifically the section "Camera Dependent ISO Definitions": > So let's say you have lenses that on each of the above 3 cameras that deliver the same photons per second to a pixel. Then if we set the exposure time the same on all three, each camera gets the same true exposure. Since each of the three cameras have differently sized pixels, the amount of light per area is different, doesn't sound like the same exposure to me, nor does it seem like a reasonable comparison. I guess if you want comparisons that make sense at 100% zoom, but why would you want that?

I think his definition of “true exposure” departs from focal plane exposure

we might be able to find his definition somewhere else on the site

oh, yeah, here it is: https://clarkvision.com/articles/exposure/

> _**True exposure**_: the actual amount of light recorded, e.g. expressed in photons (or photoelectrons) on the subject. Remember, the subject is what we are concerned about in our images, not pixels.

That sounds like light per solid angle or something, which also wouldn't result in the same amount of photons per pixel.

Other than the case where a smaller sensor and a larger sensor have an equal number of pixels, and lenses chosen to capture the same angle of view, but that wasn't the case in the ISO article. (16, 18, and 21 MPix)

High noise is there because he is using an incorrect noise model in gcam.

same noise if I use the default camera app

https://www.dpreview.com/news/2325465902/laowa-cf-12-24-f5p6-zoom-shift-lens-global-availability a shift zoom??

👀

whoaaaaaa

didn't know what shift lenses were but this sounds super cool

“keystone effect” (https://en.wikipedia.org/wiki/Keystone_effect) is maybe a better term than “perspective distortion”, given that the perspective without shift is exactly as you would expect

Dang it, Canon makes no ~18 mm, low light APS-C lenses...

Anything faster than F3.0 is for full frame cameras, making them not very wide angle on my camera

really? huh

Isn't F3 on APS-C equivalent to like F4.5 on full frame at equal angle's of view in terms of light? Kinda crazy that they don't have anything brighter than that.

Hmm, I didn't think of that

That just makes it worse

The fastest my kit lens can go is 18 mm F4.5 which as it turns out really isn't that good for nighttime photography

I’m not sure I understand the issue – it should be just as wide as an APS-C lens of the same focal length?

e.g. what’s wrong with https://www.dpreview.com/products/canon/lenses/canon_rf_16_2p8_stm ?

On an APS-C camera that becomes a 26 mm F4.5 full-frame equivalent

sure, as would any 16mm lens, whether it’s made for APS-C or not

Outside of very small cameras, like phones, all lenses are specified by their real physical focal length

So an 18mm lens gives the same angle of view on APS-C regardless of whether the lens was designed for APS-C or for full frame.

that it’s made for FF just means the image circle extends further than necessary beyond an APS-C sensor

with content that an APS-C lens would just crop off

Ah, gotcha. But on Canon's RF-S lenses (specifically for APS-C, unlike their RF lenses) don't they compensate those reported specs? Like, an RF18mm lens on a full frame camera would produce the same image as an RF-S18mm lens on an APS-C camera?

I'm 99% sure RF-S lenses have their real physical focal length as the reported focal length

> This ultra-wide-angle 10-18mm (16-29mm equivalent) Ah yeah, looks like it

That's good to know, thanks

So my F4.5 is effectively an F7.2 as well in terms of incoming light

in the Micro Four Thirds system, Olympus sometimes reports equivalent focal lengths, but they do it _only_ for the focal length and not the aperture, which is arguably somewhat dishonest (it leads many to think that they have a 600mm f/4 equivalent)

but the convention is to report the actual focal length and let people calculate 35mm-equivalent values if they need them

in some way, the “F4.0” notation kind of plays into the confusion

the “correct” notation is “f/4”, reflecting the fact that then the entrance pupil diameter is equal to the focal length (“f”) divided by 4

knowing this makes it more obvious that if the focal length is doubled, but the entrance pupil stays the same size, then the new correct expression for its diameter is “f/8”

Hmm... the total light hitting the sensor is smaller because the sensor is smaller, but light per sensor area remains the same. So with the same 300mm f/4.0 lens, a full-frame picture which is manually cropped to be 600 mm equivalent should look just as bright as an image taken on an APS-C sensor with otherwise equivalent specs (like pixel size, light sensitivity etc), no? So together with the 600 mm equivalent focal length, shouldn't f/4.0 be more representative than f/8.0?

sure, cropping is the same whether it’s done by using a smaller sensor or later in processing, but if, instead of cropping, you use a correspondingly (2× in the case of Micro Four Thirds -> FF) longer lens (i.e. “equivalent” rather than equal focal length) to still fill the larger sensor, then f/8 (and a correspondingly higher ISO setting) is what gives you similar results because then the entrance pupil (virtual image of the aperture stop as seen from the front of the lens, i.e. virtual hole through which light enters the lens) is the same size, so the same amount of light is captured from a given solid angle

so 300mm f/4 on MFT is like cropping a 300mm f/4 on FF, _or_ like using a 600mm f/8 on a full FF sensor

(or any linear interpolation thereof)

the same amount of light enters from the relevant angle of view, it’s just spread onto different surface areas, hence the different focal plane exposures

but the former is more relevant to the final results than the latter

the latter is arguably more of an implementation detail

So the 600mm f/8 on FF would appear darker than non-cropped 300mm f/4 on FF, right?

Because the physical iris size is the same, you're just sampling a smaller angle for light, making the image darker, and that's reflected by the higher f-number, as that's focal length over iris size

I think I get it

Honestly I might go for the RF24mm F1.8 MACRO IS STM. That extra wide aperture might be worth it for the superior low light performance. Since the RF16mm is also not *that much* wider than the 18mm my current lens does, and a proper low light lens is what I really want. And the image stabilization is going to allow for even better handheld light shots, compared to the 16mm F2.8 with no stabilization (I don't have IBIS)

I think its better to show the actual aperture instead of equivalent since its only equivalent for depth of field and not light right

it is equivalent for light (the same amount of light comes in from the relevant angle of view), just not light density on the sensor

(and ISO is defined in terms of light density, so indeed, despite the equal amount of light, you would use different ISO settings)

https://www.dpreview.com/articles/2666934640/what-is-equivalence-and-why-should-i-care/2

yeah but its light density thats the imporant thing

hardly

why would it be?

???

why would it not be?

if you shoot a photo with f5.6 at a 300mm equivalent FOV and 1/100 shutter

i can get a very similiar photo

if we shoot on two different formats, it will be more similar if you follow equivalence than if you target the same focal plane exposure, as described here: https://doi.org/10.1117/1.OE.57.11.110801

> In Sec. 1.1, it was pointed out that the same camera exposure used on different formats leads to images with different levels of noise and diffraction softening, the advantage belonging to the larger format. By contrast, equivalent photos have the following properties: > - total image noise of the same order of magnitude and > - same level of diffraction softening > As proven in Sec. 2, these properties arise from the fact that equivalent photos are produced by using the same lens entrance pupil diameter on each format instead of the same camera exposure.

yes but thats not relevant for the context olympus is using

it’s arguably as relevant as “you would be using 600mm to achieve similar results on FF” 🤷

they pick and choose the parts of equivalence that make them look better

I mean it's not them picking it, it sort of is convention

a misleading convention, although indeed they’re not the only ones doing it – here is Apple: https://www.dpreview.com/files/p/articles/6780391159/Apple_California_Streaming_00065.jpeg

yes everyone else does it

I don't understand the M43 market

Most of the cameras are just as big and heavy as typical FF cameras, while not being particularly cheaper than low end FF cameras.

But everyone I've stumbled on online who uses M43, uses it because of compactness. On DPReview, I only see a single compact M43 camera released in the last 5 years (Olympus PEN E-P7, which doesn't have a viewfinder). And it's not a model I've seen people mention as the one they use, those are all 2018 or earlier, like Lumix DC-GX9.

yeah, a lot of M43 users lament the direction the system has taken

I used Olympus 2019-2020 for the excellent IBIS, relatively fast readout for a non-stacked sensor, decent weather sealing and lens selection, but then moved to Canon FF in late 2020

technically, I still _have_ the Olympus gear, as I’ve procrastinated reselling it

but I haven’t used it in a while

(I really should get around to doing that)

Since it'll be a while before I can afford a modern fancy camera, I've contemplated buying something like a Sony NEX-5 and putting a pancake lens on it 😄

my dad has an Olympus. I took a look at it last week, and I was surprised by how not much smaller it is than my old DSLR.

heh, maybe if I find one of the older compact M43 cameras used at a good price, I could get one of those and borrow my dad's lenses 🙂

idk the OM systems cameras are for sure smaller than my z6 II which isnt big

bit also the lenses are much smaller as well

and for wildlife when you need big zooms that makes a difference

also for wildlife you are going ot be cropping a lot anway

Thought I'd post this here and see what results other people have been getting, since 12%, 20% and 60% are quite different numbers... https://www.reddit.com/r/jpegxl/comments/1gqd03r/comment/lwzeycr/

Nearly

I came across other red benches along the river that really reminded me of it 😁

I should really do a photoshoot around London sometime. I go there every month, get home, then realise I have a picture of a pint glass and that's it xD

nice parks there https://sami.photo/2022-10-Londres.html https://sami.photo/2023-05-Londres.html (the second album is mainly the Zoo but also a bit of Richmond Park)

I've actually never been to the zoo, but have been to the park briefly. Was giving a Canadian friend a tour around but we'd already been walking for hours in 32c heat at that point. Did go to Mudchute Farm, nice bit of green surrounded by all the glass of Canary Wharf

It's not about camera size, it's about lens size imo. An Oly 40-150 f2.8 is significantly smaller than an 80-300 f4 equivalent full frame lens

It's both the lens sizes and the feature set that the main MFT cameras provide, though admittedly other camera manufacturers have caught up on the video and IBIS front

My G9 is ironically gigantic, but my 40-150 is too useful and small to find any equivalent

strictly speaking, though, f/5.6 is enough to be equivalent

and then the size advantage is not so clear https://www.dpreview.com/products/compare/side-by-side?products=olympus_m_40-150_2p8_pro&products=sony_fe_70-300_4p5-5p6_g_oss&sortDir=ascending

the thing is mainly that in some cases, they don’t make an equivalent at all

e.g. there is no 70-200mm f/5.6 full-frame lens, as an equivalent for the 35-100mm f/2.8 for M43 (to say nothing of the 35-100mm f/4-5.6)

left to right: 70-200mm f/2.8, 70-200mm f/4, 70-200mm f/5.6 equivalent, 70-200mm f/8-11 equivalent, shedding some size and weight with each step (and, in this specific case, a clean FF / M43 divide in the middle) https://www.dpreview.com/products/compare/side-by-side?products=canon_rf_70-200_2p8l_is_usm&products=canon_rf_70-200_4l_is_usm&products=panasonic_35-100_2p8_iii&products=panasonic_35-100_4-5p6_ois&sortDir=ascending

(or in M43-equivalent terms: 35-100mm f/1.4, 35-100mm f/2, 35-100mm f/2.8, 35-100mm f/4-5.6)

decided to try and see whether this photo benefits from HDR

I think it does

(HLG image compressed using jpegli, although I forgot to make it P3 instead of Rec. 2020)

not too fond of how the SDR fallback looks, though

Not terrible, a little washed out but tapping it on mobile thankfully works

my previous, SDR-only edit of this photo was:

maybe I could brighten up the deer a bit

Both look pretty nice

Touching on this again... I could've sworn I was only getting around 20% savings at most for lossless DNG, but every file I've tried now is 50% or more

are they compressed at all in the first place?

I ran through some of my old DNGs to try out TinyDNG and found a few that went from about 25MB to 8MB or so

I think those were from my Galaxy S8, probably using Open Camera

My phone takes uncompressed 24MB files, usually gets to about 8MB when using the old JPEG compression, but TinyDNG is hitting 4MB now

Those numbers were compared to old lossless compression

Hmm... Guess I'll have to run the tests again, this is all I can find https://discord.com/channels/794206087879852103/803574970180829194/1186903859415760986

HDR photo looks nice. SDR boring.

Huh... I didn't expect that when opening it in Chrome

do you have one HDR and one SDR monitor?

Nope, only a single SDR

seems Chrome might be a bit confused about it and forgetting to tone map

(oops, forgot to make the greyscale one 4:1:0)

you have been spotted by the bird, take care

nice photos!

A road

Free to use https://commons.wikimedia.org/wiki/File:TW_CHW105_Dec_2024.jpg

full of garbare

Thanks Discord