DPReview TV: Why lower resolution sensors are not better in low light

And what happens when you combine the noisier small pixels (as you do when you scale the image to show it at the same size as the lower pixel-count camera)?

Here's a hint: it ends up being the same. If photon shot noise were the only source of noise, there would be no difference at all between pixel counts when viewed at the same size.

They made a good effort but where are the numbers, the measurements of noise to back up their observation comparing just two photos?
Unless you have abandoned your allegiance to science you have the software to measure noise in a conclusive manner. All the current test proves is a 12mp camera will show noise long before a 46 mp camera.
And where does the noise occur: shadows? Mids? What kind is it, color or luminance noise?
If you’re going to make an assertion that contradicts what is written in digital imaging textbooks and scientific papers you need more than a single very subjective anecdote, you must present evidence in the form of numbers.

The underlying principles are covered here. It's true, the video could have mentioned that fewer pixels (rather than larger pixels) will have a noise benefit at very high ISOs, where the large amounts of amplification or multiplication will eventually reveal any tiny differences in read noise.

But ultimately it's a video that's trying to be accessible to a large number of people, not a scientific paper. Hence it focuses on the impact that most people will see in most situations.

But it leaves them with an incorrect understanding; the only reason you see more noise from the A7S is because you’ve violated the design envelope of a 12mp camera. A 12mp image will produce a very nice 11x14 but that’s it. Enlarging it further is like driving 150 mph in a Morris Minor, it’s possible but it won’t be pretty.

Can you turn down the size of the files your a7r is writing to the card?

Enlarging pixels doesn't add noise, per se, it just makes the existing noise easier to see. However, combining pixels (or printing them so small that the eye/brain combines them) reduces noise.

The point is: both sensors experienced and captured the same amount of photon shot noise (randomness of light), they just did so I'm different sized buckets). Print or view them at the same size and you just see that same overall noise, regardless of the size of the capture buckets.

The differences start to appear at very high ISOs where the sensor with fewer pixels (and fewer pixel read events) has fractionally less (electronic) read noise. The large amounts of amplification/multiplication start to exaggerate/reveal these differences.

Dunning-Kruger Effect - https://en.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect

In these comments I see a few very vocal ignoramuses who have declared, with zero evidence, that this very reasonable video is wrong and they are right. They do not know they are wrong and will not accept that they may be wrong. A case study of the DK effect.

It's so funny when geeks find a scientific psychological term with a fancy name they can throw in the face of naysayers.

"They’re just showing that the more megapixels you have the larger you can print without seeing noise. How is that news?" It's apparently news to the few, very vocal, people who continue to dispute the results despite seeing it plainly in the video.

@ yakovlev: You forget that when you have larger photosites on a given surface area, you also have fever of them.

When you mention extreme low-light scenarios, I assume you are talking about using data close to the noise floor. For real world use - also astrophotography, except for very specialized work when you use specialized cooled cameras anyway - we never use such bad data. How often do people use ISO values above 200 000 for their real world photography do you think?

5-10% differences in noice level is hardly perceivable.

If you compare prints if different sizes it wouldn't make any sense...

I'm talking about the print resolution not the size. I know they are the same size but I think it was mentioned that they were printed at different dpi. I'd like to know why. I know nothing about professional printing. Are you saying they had to use different dpi to get the same size because of the different resolution of the sensor?

Ah, I believe they meant the resolution that went into the print, not the resolution of the print.

BobT, as they said, they did no interpolation (up-scaling) before printing. The resulting dpi that they're talking about is simply the native resolution of the cameras divided by the size of the print.

However, of course the printer itself only prints at one dpi, probably 300. So the printer has to do the interpolation. The advantage of doing the interpolation on a computer before printing is that you can then judiciously apply sharpening to make the interpolation look better and more crisp. However you can't *create* more detail by doing so.

@Richard Butler

> Read noise differences start to add up at very high ISOs
> and in very long exposure photography, for instance.

Sony IMX455 sensor (Sony a7R IV) and its MF and ACS-C/1" variants (e.g. IMX571 in 26Mp Fuji cameras) have incredibly low read noise, about 1e- per pixel. So it gets swamped by shot and thermal (a.k.a. dark current) noise, especially during long exposures.

Interestingly, the advent of IMX455-based astro cameras (QHY600, ASI6200) caused revolution in amateur astrophotography. Until then, the conventional wisdom had always been that bigger pixels (of sensors like Kodak KAF16803, 9um!) and longer subexposures were always better. Well, not anymore: even though pixels of IMX455 are almost 6 times as small (by area) as those of KAF16803, read noise of the former is so small that people can now use many times *shorter* subexposures.

@Richard
Medium Format does look better, because it does have a bigger Sensor(or bigger Film format size), bigger Pixels, and a different kind of Rendition, DoF, in terms of Bokeh, Tonality, Transitions, Textures, etc. in contrast to "Fullframe" 36x24mm, i'd like to add.

Good Light.

vadims, you pick exact *one* possibility, maybe thing out of it. When i speak from MF, i mean usually Film. Into 2nd place digital. And MF does have for real all the things, written above.

There is a specific medium format look, one won't have with 36x24mm, it looks differently.

Good Light.

@marc petzold

Marc, we're veering more and more off course of the OP, but anyway... :-)

> There is a specific medium format look, one won't have with 36x24mm

... until perspective/framing and DOF are matched.

Michael Reichmann, who was an avid MF shooter, once shot an image with an MF camera (can't remember which, he had plenty) stopped down, and a point-and-shoot (!) camera; printed both, and hung them in his gallery. He then did a blind test with visiting fellow photographers, asking which print was made with which camera (telling them in advance the options). The guess rate was approximately 50/50. There was an article at LuLa about that.

BTW, right below this post kodacolor200 wrote: "The EM1 destroys my current gen BSI FF cameras".

Everyone believes what they want...

Yes - that is the reason Nikon professional camera D5/6 only has 21 MP whereas D850 has 46 MP and Sony Vlog camera a7S III comes with a 12 MP full frame sensor

Bill is a good guy and I prefer his site to DxO. There is some other reason then why the GH5 mark II is virtually equal to recent APSC cameras and a 1/3rd of a stop better than the GH5. Both DxO and Bill found the same results. APSC has an advantage at iSO 100 and the GH5II has one at high (near unusable) ISOs.

https://www.photonstophotos.net/Charts/PDR.htm#Panasonic%20Lumix%20DC-GH5M2,Sony%20ILCE-6600

Mike, LOL...I can just imagine the idea of "the image sensor of your existing camera and it would look like you were looking at yourself in a mirror."

Note: I'm not laughing at you but with you...that's a pretty funny idea.

Imagine the marketing for such a camera: It's the ultimate selfie camera! No need to even take the image, or use a lens, just take the lens cap off and admire yourself in the image sensor. No EVF lag, no buffering issues, no storage card needed. Just stare in delight.

Anyway, thanks for the humor...

It wasn’t meant to be funny.
What’s funny is the ridiculous idea that you can get a further 1 stop advantage (I.e, double the light) with a new AR coating.

Again the implication is that existing sensor AR coatings are so inefficient that they reflect half of the light back through the lens.

So silly. But some people believe the marketing.

An M43 sensor performing like an APS-C sensor means 1 stop. Unless your APS-C is the smaller Canon APS-C in which case it’s closer to 0.7 stops.

So you have no clue what you are talking about. That helps.

M43: 17.3x13mm, 225mm²
Canon APS-C: 22.2x14.8mm, 329mm²

Horizontal crop factor: 22.2/17.3 = 1.28
Vertical crop factor: 14.8/13 = 1.14
Diagonal crop factor: 28.3/21.6 = 1.23
By area: log₂(329/225) = 0.55 stops

Regular APS-C: 23.6x15.7mm, 370mm²

Horizontal crop factor: 23.6/17.3 = 1.36
Vertical crop factor: 15.7/13 = 1.21
Diagonal crop factor: 28.3/21.6 = 1.31
By area: log₂(370/225) = 0.72 stops

So again you are wrong, its .72 stops and .55 stops.

The GH5 was already out performing the A6600 because it was well within .72 stops. And the GH5 II, if you can believe P2P, erased the equivalence gap.
Now I wonder if the coating did help some.

You neglected the difference in aspect ratio in all of your calculations. Please try again.

I also don’t know where you get that the gH5 is well within 0.72 stops (again your math is wrong but that’s another story.)

To me these two look quite similar in noise and shocker, a rad bit more detail available on the (slightly) higher resolution sensor. (Consistent with the video although I’m not saying for sure that the small resolution increase is the cause.). Note GH5 is iso 6400 and a6600 is iso 12,800.

https://www.dpreview.com/reviews/image-comparison/fullscreen?attr18=lowlight&attr13_0=panasonic_dcgh5&attr13_1=sony_a6600&attr15_0=raw&attr15_1=raw&attr16_0=6400&attr16_1=12800&normalization=compare&widget=585&x=0.33276826181646946&y=-0.07012029150935115

"You neglected the difference in aspect ratio in all of your calculations"

You actually think aspect ratio affects the area of a rectangular sensor and how much light it gathers!
I wish I saw the look on your math teachers face when you told him the 3x10 rectangle had more area than the 6x5 one.
Or when you told everyone you would need more paint for a 5x12 wall than a 6x10 wall.
I hope everyone in the Photography Science and Tech forum sees this, They'll enjoy a good laugh too.
Mike thinks the most poplar aspect ratio (M43, MF, smartphone, etc) for camera sensors somehow makes them gather less light!

The Sony definitely looks worse at 6400 to me.

As far as the photons to photons graph, it's very possible that the Panasonic is shifted, where the iso values don't mean the same thing as the Sony. It's clear the Sony has better absolute performance from that plot.

And no, apsc does not have a one stop advantage over m43. You need a sqrt(2) crop factor for that. Full frame does have significantly more than a one stop advantage over apsc though...

Y/N are you going to tell the whole world that they are wrong when you talk about crop factors? Crop factor for M43 is 2.0x no? Why do you think that is?

Anyway not gonna waste another breath on you. Have fun in your imaginary land.

Its good to see everyone didn't fall for Mike's misinformation.
Its math and there is no debate here.

And Mike the discussion is about M43 vs APSC and the difference in stops.
"Crop factor for M43 is 2.0x no? "
No, not between APSC and M43, which is what we need to determine the difference in stops between APSC and M43.
Remember, the topic is about how the GH5II over comes the ,72 stop difference between M43 and APSC.

M43 Crop factor is 2.0
APS-C crop factor (non Canon) is 1.5
APS-C crop factor for Canon is 1.6.
The reference is and has always been FF.

These are facts.

If you want to compare crop factor for APS-C (non Canon) vs M43 that’s fine. It’s 1 1/3x. And if you want to convert that to a stop advantage it’s 0.83 stops. Sure. A little short of 1 stop.

My statement that you can gain a 1 stop advantage (or a 0.83 stop advantage or a 0.7 stop advantage. Or a 0.5 stop advantage) due to some new revolutionary AR coating is just as ridiculous now as it was when I made my initial comment.

Mike, you are all over the place, (still wrong, but getting closer)
I even did the math for you above, lol.
Diagonal crop factor: 28.3/21.6 = 1.31 (that you finally got real close)

As for the statement, it is pretty amazing the GH5II sensor virtually equals the A6600 in DR (at some ISOs the GH5II is better). And that is according to both DxO and Photons2 Photos.
It will be interesting to see of the GH6 passes the A6600. (it will for video, IBIS, and everything else of course).

What’s interesting is you said previously you preferred a 2:1 aspect ratio but then you insist on using this (useless in your case) 4:3 diagonal to try to give m43 an advantage.

Just proves you are here to troll.
Take care.

Again, I am sorry for mentioning that other brand when talking about the GH5 II and its sensor improvements. I should have said Nikon or Canon, and then it would have stayed civil.

The summary from that article doesn’t do justice to the 4500 words he wrote. If you ask me his summary statement is misleading.

The key is actually in the headline. He basically writes 4500 words to explain why sensor size matters. The article isn’t called “why pixel size matters”

DPReview TV: Why lower resolution sensors are not better in low light is misleading. I stated I wanted an updated view from folks like Roger Cicala seeing how we are now in 2021 an the advances since then. Roger covers pixels and sensor size, folks will make out of it what they want, as they always do.

of course sensor size does matter for noise and low-light performance. In fact this is the thing that matters most. Just don't understand how somebody could write such a nonsense that it does not matter )

It’s a summary bullet taken out of context of the whole 4500 word article. Not taken out of context by BTN. Taken out of context by Roger himself. Roger’s summary doesn’t summarize his article effectively if you ask me.

The article is fine. The summary is not.

BackToNature1 - It's not a question of advances in tech, it's more about changes in perspective. Roger's article seems to take a pixel-level approach (and at first glance, possibly focuses a bit too much on the impact of read noise, rather than also considering shot noise). Chris and Jordan are talking about the effect on your overall image, not the individual pixels (because, when you present images at the same size, the pixel level noise stops being the thing that matters).

Also "Both cameras were shot in the highest quality .JPEG", and they talk about sharpness...

War is peace. Freedom is slavery. Less is more?

it's actually a pretty good video. The point being made is not that less MP is better, but that higher MP may not matter that much for most uses.

He does address one reality that many overlook, and that is the fact that most photos will be viewed on the electronic screen. Even his test on Instagram is measuring a real world situation for many.

Now with the prints, I don't see in the video where he says these are not high quality prints, like tozz claims. In fact he gives a promotional shoutout to the printer company, and I don't think he would do that if he said they are making low quality prints.

But again, his point is that higher MP may not matter as much as many may think it does, not that it doesn't matter at all.

BTW, his test is in interesting one. I bet most people, if they saw images taken with the higher and lower res cameras, couldn't tell the difference.

It’s in the end of the video (of course) where he sums things up. Anyone who has printed from high resolution cameras know it’s not an easy thing to get right, even with my Z7 I had major artifacts using normal poster printing when disabling the printer lab optimizations (the way they avoid that is… blurring). In the end I went with the higher quality printers and all the false details disappeared while keeping the real, as captured.

The video doesn’t even mention PPI nor the service used (and what configuration), again, everyone who does prints from high resolution cameras should be pretty aware of the importance of this.

Again, it’s fine if he doesn’t need more than 12mpix but he’s spreading a lot of confusion (just how people who claimed lower res cameras had lower noise did/keep doing)

A properly printed photo from the 102mpix would be easily distinguishable from 12mpix at A3 or larger at the distance they’re looking. But again, printing takes knowledge, it’s more than uploading a jpeg on a website.

Cool that I started a discussion at least.

Nerds are consumers and consumers are nerds.

And nerds are cool?

"[...] nine large buckets [...], they're going to collect more water than the same area occupied by 81 buckets [9x9] because of the rims of the buckets. [...]

Very true, but this limitation has been reduced to almost negligible levels many years back, since the gapless micro lenses and other improvement in manufacturing. And since BSI it is obsoleted completely.

If anything more megapixels actually gives better color information to start with, since it subsamples more RGB values on the same area (not discussing detail at all).

@jonas ar have you seen the big print in the video - we are talking about thus size.

I understood that cosinaphile suggested to compare images from a 12MP and a 61MP camera at 100%. This will result in analysing vastly different parts of the scene and will not provide a sensible comparison. The images should be analysed at the same size to reveal their relative merits. It does not matter how big as long as they are the same

In fairness to cosinaphile, a difference does start to appear at super-high ISOs.

This article which Chris and Jordan refer to in their video includes some of the maths explaining why the increased number of read noise events in a higher pixel-count sensor can eventually see it drop behind (even though the amount of read noise per pixel is lower).

The question for each photographer is: how often do I use ISO >102,400 and is the difference big enough to give up the detail benefit of the higher pixel count at lower ISOs (the answer will vary depending on what/how you shoot).

This video is compatible with equivalence if you actually understand them; in fact this article on equivalence (which contains a section on smartphones) makes a similar point as the video, in section 4.3: https://doi.org/10.1117/1.OE.57.11.110801

cosinaphile, learn spelling and grammar before criticizing anyone else's work or intelligence. Internet arguing 101, really.

I have a 24 megapixel 2012 Nikon D600 (DSLR) and a 12 megapixel 2012 Nikon P7800 (digicam). Both are fine for my needs, but the P7800 Is unusable at ISO 800, while the D600 does credible work for me at ISO 6400. Sensor (pixel) size, perhaps?

Your d600 has 20x the area of your p7800 (about a 4.5x crop factor).
Your ISO values seem about what you would predict with this knowledge, but I would expect the p7800 to be unusable at ISO400, since that is like ISO8000 on the larger sensor... :)

There's no 'agenda' cosinaphile (I can't even think of why it would make sense for someone to take a particular angle on this issue).

It's not about prints, specifically, it's purely about whether you think a 'per image' or 'per pixel' perspective is more valuable. Because most modern sensors have extremely low levels of read noise, most of the noise we encounter comes from photon shot noise (randomness of light). At which point, overall image noise is a question of how much light you captured to make the image (dictated mainly be shutter speed, aperture and sensor size, or shutter speed and equivalent aperture, whichever way you prefer to think of it).

ie: When viewed at the same size, the images of two sensors with different pixels counts will have similar amounts of noise (because the higher amount of pixel-level noise averages out when you combine those pixels to downsize to the size of the lower pixel-count camera).

...

Because DPReview started in an era when pixel counts were insufficient for large prints, we focused on 100% crops (pixel level detail). In hindsight I think we stuck to that approach for too long, having not recognized that we'd passed a point where most cameras have (at least) sufficient pixel counts for most people's needs. At which point an image level perspective becomes more meaningful.

We still provide the images with pixel level detail, if that's more meaningful for people, but we tend to discuss images at a whole-image level.

It's exactly the effect that Chris (and many film developers) are expressing when they talk about enlargement or magnification. ie: Medium Format looks better because you haven't enlarged the image so much (to achieve whatever print size) is essentially a process-focused way of saying 'the larger capture area captured more light, so has less shot noise (better SNR) at an image level.'

spider-mario: No, we want to let people see the original image as the starting point, but all links in reviews that discuss noise should be in 'Comp' mode.

i feel it is an agenda .. ut i respect your right to think otherwise
its true under many conditions noise from different mp cameras will appear similar

but in absolute terms larger photo sites are superior .. you cannot dispute this

do you think the sensor wells in a 12 mp cell phone sensor can deliver the same quality ... yes quality as a 12 mp full frame sony full frame camera ? noise ? next time do the comparison at iso 51200 and we will see a real noise advantage

necxt time print them both out at their respective FULL sizes 12 mp vs 61 mp at high iso not 1600 and see what you get

i say that is bonkers and i have believed you have a pro cellphone agenda and have for yrs.. its simply my belief

its why you pick at nuance in things like comparing equivalency in 1 inch \m43 sensor \apsc and FF but never ever apply the thinking or numerical comparisons to cell phone censors

ever .and that , respectfully is why i believe there is an agenda

“its true under many conditions noise from different mp cameras will appear similar
but in absolute terms larger photo sites are superior .. you cannot dispute this”

Yes, you can, as I did above. It’s explained in the link.

“do you think the sensor wells in a 12 mp cell phone sensor can deliver the same quality ... yes quality as a 12 mp full frame sony full frame camera ? noise ?”

But it’s not because of pixel size. A full-frame camera with pixels of the same size as those of the smartphone (and consequently much more of them) would also be capable of much less noisy images.