Confusion between pixel size and low light performance

[SeeingSomeThingsMissingMany]

The most important factors are the lens and the photographer. A good lens at optimum focus and optimum aperture can project several hundred MP worth of detail on the sensor, according to Jim Kasson's calculations.

But, in the real world subjects are not in absolute focus and there is an acceptable Circle of Confusion, yielding some Depth of Field. There may also be some camera and subject motion and air turbulances. Stopping down, diffraction also comes into play past medium apertures.

So, the image projected on the image may not have a lot of fine detail. But, the sensor really needs to resolve all detail.


This Phase One P45+ image shows aliases on some detail, so it would need higher sensor resolution.


This Phase One P45+ image shows no obvious aliases. So it does not need more pixels. But, it may be quite possible that it may benefit from more pixels.

Best regards

Erik

I imagine these photos would evoke memories for you Erik.

Where there are lots of detail also to minimise aliasing is a reason I still have Gfx100s/ii in my distant radar.



--
Photography after all is interplay of light alongside perspective.

 

[SeeingSomeThingsMissingMany]

I don't really know whether I have any particular wants beyond enough resolution and low enough noise to print to the size I think of as my standard output size (12" x 12").

I think I acquired the gear to achieve that about back about when I bought an Olympus E-410. Everything since then has probably been an unnecessary waste of money driven by gear angst and GAS fomented on these very forums (plus the never-ending search for convenience)

Personally, if I look back at my older prints, they don't look any worse than what you get from modern cameras. You only get into trouble if you try to print too large, then modern cameras come into their own. But within the maximum size capabilities, the image quality appears to me to be more down to the printer, paper and ink than the camera. I really don't see any difference in basic image quality between any cameras I've had, going right back to my Oly E10 as long as you print small enough. I used to print A3+ with that E10 and by modern standards the print quality is awful, there simply isn't enough detail. But print at 8" x 6" and the prints look no different from my GFX or Sony. As I see it, sensor development has given us the ability to print larger and larger but things have been mostly stable in the last decade unless you need the ability to print 60" wide.

During lockdown a person approached me with regards to displaying in a gallery exhibition in london 27 of my monochrome digital captures I'd taken with 2decade or so older digital cameras.

I was going to have them printed small according to the sensor ratio photograph ratio by a specialist that instinctively knew which paper would suit which photograph.

I wanted visitors to the exhibition to be drawn in to the small prints internalise them easily. Were they to purchase to then be able to easily move it around take it with them anywhere in a bag. Also because they wouldn't hold up beyond 7" 10".

Sometimes I want some of my photographs printed 20ft big other times I prefer intimacy smaller prints.

--
Photography after all is interplay of light alongside perspective.

 

[SeeingSomeThingsMissingMany]

This led me to believe that the bigger the pixel size the better low light performance a sensor would have (sensor size being constant).

However the 50R and 100S changed that perception because the 100S is noticeably better in low light than the 50R even though the individual pixel size is smaller. Then I realized that both Sony sensors are BSI which led me to believe that it was the BSI part of the 100S that made it a better low light performer compared to the FSI sensor on the 50R/S/II since there is more area to collect the light. Is that true or is that because the 100S uses dual conversion gain?

Now the question is if there was a BSI 50MP medium format sensor would it perform better in low light compared to the BSI 100MP sensor? What if the 50MP sensor used dual- conversion gain?

There was a thread here couple months earlier a gent with his 50s or 50r had a lot of blurry pictures inside a low lit gallery/museum. It was because the Gf lens was slow and there was no IBIS on his Gfx.

For me I would take a F1.4 1" sensor compact Lx10/Lx15 for this. Latest BSI Gfx sensor becomes moot for me in such when the Gf lens is slow.

 

[Manzur Fahim]

If you downscale the A7R II (42mp) image to A7 III (24mp) level, you'll see that the noise floor is decreased because you are pushing the pixels together.

Before everyone used to think that larger pixels are better because they have a large surface area to capture light. But if you have for example, 4 small pixels which takes the same surface area as one large pixel, then combined, the 4 small pixels is capturing the same amount of light. This is what my understanding is from some articles, someone else might be able to explain it better.

I wanted fat pixels of A7s 12MP, but then I wanted detail of A7r 36MP. Which to choose. Downsampling A7r to 12MP would give me both so for me A7r.

Imagine what the GFX 100MP sensor can do then

 

[Scottelly]

I can follow the logic of what is being said, and therefore if I want the best DR, colour fidelity and smoother tonal transitions I should bin my 50sii for a second hand 100s, prices being more or less equal now.

All those parameters may have no relevance at all.

1. DR small difference
2. Colour fidelite is exactly the same
3. Tonal transitions, please define, but exactly same except deepest darks.

But not to me and its hard to define tonal transitions beyond emotive language because it will very much depend on things that are probably not measurable such as smoothness of transitions from one tone to another, how soft/smooth the colours look and so forth - which is why I'm pretty sure it's not a measurable parameter, but more a 'feel', which of course is an anathema to a gear site

But, would I actually see the difference in a print?

The 50 range will print A1 at 249ppi

The 100 range will print A1 at 352ppi

Both will be resampled to 360 PPI, at least on Epson printers.

Which means no, as I've played the game of resampling 50 images to 100 sized images based on the DPR test images and beyond a colour difference in the original images, didn't reveal any startling differencess

Having read all the posts, will anyone discern a difference between either in a long exposure image. Detail isn't the thing I'm really interested in as the 50 has plenty enough of that for me, its all about colour, tonality and transition between tones.

Genuinely interested as whilst I enjoy reading these discussions, I'm somewhat sceptical when it comes to real world imagery. The phrase 'marginal gains' comes to mind in that there may be an incremental difference in theory, but it isn't necessarily realisable in print.

The most important factors are the lens and the photographer. A good lens at optimum focus and optimum aperture can project several hundred MP worth of detail on the sensor, according to Jim Kasson's calculations.

Would agree - it becomes more a factor of focusing, exposure intent and so forth

But, in the real world subjects are not in absolute focus and there is an acceptable Circle of Confusion, yielding some Depth of Field. There may also be some camera and subject motion and air turbulances. Stopping down, diffraction also comes into play past medium apertures.

Again, that's about sharpness not colour, tonality etc

So, the image projected on the image may not have a lot of fine detail. But, the sensor really needs to resolve all detail.


This Phase One P45+ image shows aliases on some detail, so it would need higher sensor resolution.


This Phase One P45+ image shows no obvious aliases. So it does not need more pixels. But, it may be quite possible that it may benefit from more pixels.

Best regards

Erik

Your thoughts gents

Andy

I'll say it again, it's not all about the sharpness for me at least. Some of my favourite images have pretty much no detail but absolutely tons of colour and tones

This tonality, colour and tone stuff is mostly emotive hot air that wouldn't stand up to a double blind test in my opinion.

Back in the film days, larger formats had smoother tones than 35mm because they were enlarged less for the same print size and so the grain was less prominent. It's the grain that destroyed the smoothness of tonality.

Digital doesn't really have grain, only digital noise, and on modern sensors noise is extremely low if properly exposed at base ISO. I have never seen any visible difference is tonality/smoothness on any high quality digital image that couldn't be put down purely to post processing.

Maybe if you making prints metres wide something will show up, but otherwise I doubt it.

DXOmark does measure colour bit depth which they say measures the numbers of colour transitions until obscured by noise (see the impact of grain/noise on tonality once again!). They say differences of less than 1 bit are essentially invisible. I sorted their sensor league table by colour bit depth. The best sensors (maybe the top 50 cameras in their database) differ by less than 1 bit.

Issues around colour discrimination, tonality and so on, are largely issues about noise. Reduce noise to a minimum and what is left is the imagination of the photographer and some nostalgia about medium format and large format film.

Aside from one P1 sensor, the best performer in the dxomark database is the Lumix S1R. Not really a camera being trumpeted far and wide by reviewers for its wonderful colour and tonality.

It's all audiophile style nonsense and imagination. The 'medium format look', the 'Leica look', the 'Zeiss look', 'tonality'. Just take a good photo and no one can see these 'looks' under blind conditions.

David you may have just convinced me to get an S1R. I can use all my Sigma lenses on it if I do. I wonder if I would prefer it to my giant Fuji GFX100.

--
Scott Barton Kennelly
https://www.bigprintphotos.com
https://www.sigmaphotopro.com
https://www.sigmacamerapro.com

 

[Erik Kaffehr]

I have noticed that many confuse a larger pixel size (photosite) with better low light performance compared to a higher resolution but smaller pixel size sensor. I am guilty of the same. Coming from the Sony A7 universe and A7RII being my first digital camera I read all about low light performance of the Sony FF cameras and from what I saw the A7III had a larger pixel size but better low light performance. Since then I switched to the A7III for astro. This led me to believe that the bigger the pixel size the better low light performance a sensor would have (sensor size being constant).

However the 50R and 100S changed that perception because the 100S is noticeably better in low light than the 50R even though the individual pixel size is smaller. Then I realized that both Sony sensors are BSI which led me to believe that it was the BSI part of the 100S that made it a better low light performer compared to the FSI sensor on the 50R/S/II since there is more area to collect the light. Is that true or is that because the 100S uses dual conversion gain?

Now the question is if there was a BSI 50MP medium format sensor would it perform better in low light compared to the BSI 100MP sensor? What if the 50MP sensor used dual- conversion gain?

Now to add to this confusion is the fact the both Sony sensors have almost identical PDR-ISO curves yet the A7III is noticeably better in low light compared to the A7RII. Why is that?

If you downscale the A7R II (42mp) image to A7 III (24mp) level, you'll see that the noise floor is decreased because you are pushing the pixels together.

Before everyone used to think that larger pixels are better because they have a large surface area to capture light. But if you have for example, 4 small pixels which takes the same surface area as one large pixel, then combined, the 4 small pixels is capturing the same amount of light. This is what my understanding is from some articles, someone else might be able to explain it better.

Hi Manzur,

Your writing applies to Signal Noise Ratio at high and medium exposures, but darks are a bit different.

To that comes the area of the wiring and the transistors handling the photodiode. If we have a given set of design rules, the wiring/junction area is constant for the pixels, so making them smaller reduces the photodiode area.

But, I think that smaller pixels use tighter design rules, which may need a different process, possibly needing a newer 'fab'.

BSI design moves the wiring behind the photodiode. I don't know about the transistors.

Photographics sensor are said to be some of the most complex to design analogue devices.

In essence, you are essentially right but it may be a bit more complex than that.

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic tends to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Peter has left]

Aside from one P1 sensor, the best performer in the dxomark database is the Lumix S1R. Not really a camera being trumpeted far and wide by reviewers for its wonderful colour and tonality.

It's all audiophile style nonsense and imagination. The 'medium format look', the 'Leica look', the 'Zeiss look', 'tonality'. Just take a good photo and no one can see these 'looks' under blind conditions.

David you may have just convinced me to get an S1R. I can use all my Sigma lenses on it if I do. I wonder if I would prefer it to my giant Fuji GFX100.

Hi Scott!
If you have a lot of first-class Sigma lenses, then the Panasonic S1R (with pixel shift) is a good choice, but for regular photography it has few MP.. Your Fuji GFX 100 is definitely the best for that.

If you want, I'll add a photo of the size here as a sample 24x36 inch size - 7092x10630px/300ppi.

 

[Manzur Fahim]

David you may have just convinced me to get an S1R. I can use all my Sigma lenses on it if I do. I wonder if I would prefer it to my giant Fuji GFX100.

Another camera that I want to buy, but not available where I live. I like Panasonic colors and tones; they always seem so neutral and natural.

Another one that got away.

 

[WhiteCranePhoto]

I have noticed that many confuse a larger pixel size (photosite) with better low light performance compared to a higher resolution but smaller pixel size sensor. I am guilty of the same. Coming from the Sony A7 universe and A7RII being my first digital camera I read all about low light performance of the Sony FF cameras and from what I saw the A7III had a larger pixel size but better low light performance. Since then I switched to the A7III for astro. This led me to believe that the bigger the pixel size the better low light performance a sensor would have (sensor size being constant).

However the 50R and 100S changed that perception because the 100S is noticeably better in low light than the 50R even though the individual pixel size is smaller. Then I realized that both Sony sensors are BSI which led me to believe that it was the BSI part of the 100S that made it a better low light performer compared to the FSI sensor on the 50R/S/II since there is more area to collect the light. Is that true or is that because the 100S uses dual conversion gain?

There's a lot involved. BSI does improve low light performance because the wiring in front of the photo sites does shadow them.

But there are quite a few other factors, including heat. There will be more noise in a hot sensor than a cool one, and more noise means poorer low light performance. Because of that, improving heat dissipation and reducing power draw (and therefore heat) from the camera's processors also helps to improve low light performance.

If all else is equal, then bigger photo sites will lead to better low light performance, and BSI will lead to better low light performance than FSI, etc. But so does dual gain analog to digital conversion and coatings that improve light transmission on the sensor itself.

 

[JimKasson]

I have noticed that many confuse a larger pixel size (photosite) with better low light performance compared to a higher resolution but smaller pixel size sensor. I am guilty of the same. Coming from the Sony A7 universe and A7RII being my first digital camera I read all about low light performance of the Sony FF cameras and from what I saw the A7III had a larger pixel size but better low light performance. Since then I switched to the A7III for astro. This led me to believe that the bigger the pixel size the better low light performance a sensor would have (sensor size being constant).

However the 50R and 100S changed that perception because the 100S is noticeably better in low light than the 50R even though the individual pixel size is smaller. Then I realized that both Sony sensors are BSI which led me to believe that it was the BSI part of the 100S that made it a better low light performer compared to the FSI sensor on the 50R/S/II since there is more area to collect the light. Is that true or is that because the 100S uses dual conversion gain?

Microlenses affect this comparison.

There's a lot involved. BSI does improve low light performance because the wiring in front of the photo sites does shadow them.But there are quite a few other factors, including heat. There will be more noise in a hot sensor than a cool one, and more noise means poorer low light performance. Because of that, improving heat dissipation and reducing power draw (and therefore heat) from the camera's processors also helps to improve low light performance.

If all else is equal, then bigger photo sites will lead to better low light performance,

That, in my testing, is a marginal effect, and is more than cancelled by the fact that nonlinear noise reduction works better at lower pixel pitch, all else equal.

and BSI will lead to better low light performance than FSI, etc. But so does dual gain analog to digital conversion and coatings that improve light transmission on the sensor itself.

 

[WhiteCranePhoto]

If all else is equal, then bigger photo sites will lead to better low light performance,

That, in my testing, is a marginal effect, and is more than cancelled by the fact that nonlinear noise reduction works better at lower pixel pitch, all else equal.

Which goes to show how nearly impossible it is for all else to be equal