[normhead]

I'm sensing confusion, so here's a simple example: large pixel photo site with capacity 4 vs. small pixel sensor area covered by 4 pixels with capacity 1 each (unrealistically assuming a fill factor of 100%). Photons hit small pixel sensor area with distribution 2/1/1/0, pixel #1 clips, result after downsampling to low resolution is 3, large pixel result is 4.

A couple of example photos would clarify this really well. The first question that would come to mind would be, why would you clip the 4th pixel? Of course if you set up an artificial exclusion in your framing of the question, you can create a scenario that true or not will lend you to believe the above post. But that's only because you're too obtuse to come up with solutions as to what can be done with the 4th pixel in your example. You assume it will be wasted, and then proclaim your results, as if any engineer in his/her right mind would do that.

But I digress, what do you think of that "Flintstone's " furniture?

[ElJamoquio]

A couple of example photos would clarify this really well.

https://xkcd.com/1416/

[normhead]

https://xkcd.com/1416/

Maybe not those photos.

[Ikarus]

The first question that would come to mind would be, why would you clip the 4th pixel?

It's called a thought experiment. Full-well capacity of 1 per higher resolution pixel, 4 times that for the super-pixel covering the area of all 4 of them. Feel free to point out why this should not be comparing apples to apples.

But that's only because you're too obtuse, to come up with solutions as to what can be done with the 4th pixel in your example.

Yeah, I love you too, buddy. I could think of a few things, actually, but not all of them make for a favorable tradeoff, especially not if the maximum resolution you're interested in is the one you're already getting from the larger pixel sensor.

[normhead]

OK then, lets look at some real world examples from Imagine Resources where you can compare a 20 MP canon 6D, with a 24 MP K-3, using similar lenses. The K-3 has a noticeably smaller pixel size, in part because it's more MP and inert because the sensor is half the surface area to begin with, so the famous double whammy.

Yet it produces distinct lines down to 2400 lw.ph. Click here for reference.

The Canon 5D mk III also tops out at about 2400 lw/ph Click here for reference.

My Pentax K-3 with a very similar lens produces 2700 lw/ph in similar conditions. Click here for reference.

Thus in using much smaller pixel size on a half size sensor. I get more resolution. How does that fit with your theory?

I have to say, these are real world examples. Not some guy wandering around in the canyons of his mind.

[falconeye]

I'm sensing confusion, so here's a simple example: large pixel photo site with capacity 4 vs. small pixel sensor area covered by 4 pixels with capacity 1 each (unrealistically assuming a fill factor of 100%). Photons hit small pixel sensor area with distribution 2/1/1/0, pixel #1 clips, result after downsampling to low resolution is 3, large pixel result is 4.

Ikarus, your reply tells me you started to think over your proposition but just don't want to be wrong.

Well, that's all very human, so I'm fine with that. Although it doesn't touch at what has been wrong with your recent arguments higher up in this thread.

As for your thought experiment which I quoted: That's about correct for 1-bit pixels which is why I already accounted for in my previous thought experiment allowing for a margin to cope with noise clipping.

However, it is entirely irrelevant for real world sensors. Even high resolution cameras have well capacities of 50,000 or even 70,000 per pixel. The effect you describe therefore is less than 0.5% (the math is in textbooks about statistics) or less than 0.01 EV. This is unmeasurable and as I said, irrelevant.

[Ikarus]

OK then, lets look at some real world examples from Imagine Resources where you can compare a 20 MP canon 6D, with a 24 MP K-3, using similar lenses.

I'm not sure I understand how your example relates to the question of if and how the pixel DR affects the DR of the sensor. As far as practical examples you're asking about - I'm wondering why the 16mb K-5ii handily beats the 24mb K-3 in dynamic range by a whopping 0.7 bits in DxO's size-normalized measurements, despite having a sensor of the same size. That has nothing to do with the fact that the K-5ii has larger pixel elements? I'm not convinced.

[normhead]

Yeah, I love you too, buddy. I could think of a few things, actually, but not all of them make for a favorable tradeoff, especially not if the maximum resolution you're interested in is the one you're already getting from the larger pixel sensor.

Where in that sentence did Dynamic Range become the topic of discussion?

I really hate it when I have to tell folks what they said, because they don't even know what they said.

I shoot with both a K-5 and a K-3, and I'm well aware of the difference that .7 EV makes, and in my opinion, it's not worth changing cameras for. If I was shooting a sunset, I wouldn't change from a K-5 to K-3, but I wouldn't change from a K-3 to a K-5 either. What are you shooting with? The fact is, different sensors have different DR, and it's very nice to have, so is extra resolution.

[Ikarus]

which is why I already accounted for in my previous thought experiment allowing for a margin to cope with noise clipping.

Well, but that's the whole point now, isn't it? The example demonstrates that you need to keep less margin for preventing clipping with the larger-area pixels, which directly translates to a higher usable dynamic range.

However, it is entirely irrelevant for real world sensors. Even high resolution cameras have well capacities of 50,000 or even 70,000 per pixel. The effect you describe therefore is less than 0.5% (the math is in textbooks about statistics) or less than 0.01 EV. This is unmeasurable and as I said, irrelevant.

For the argument to hold it is not necessary to consider the FWC as the limiting factor. It still holds if the clipping happens elsewhere in the analog signal chain, as long as we ensure that for apples-to-apples comparison, the clipping threshold is at the same readout value for a given light intensity (i.e. it scales with the pixel area).

---------- Post added 07-30-15 at 06:23 PM ----------

Where in that sentence did Dynamic Range become the topic of discussion?
I really hate it when I have to tell folks what they said, because they don't even know what they said.

I really hate it when people take things out of context, or worse, jump right in going on the offensive without having bothered to read any of it. This whole discussion (the recent part of it, anyway) is about whether there is a trade-off between resolution and dynamic range. Right above the part you excerpted because you had keyword-spotted the term 'resolution', I had written:

It's called a thought experiment. Full-well capacity of 1 per higher resolution pixel, 4 times that for the super-pixel covering the area of all 4 of them. Feel free to point out why this should not be comparing apples to apples.

With the terms 'full-well capacity' and 'clipping' showing up in the thought experiment, didn't that make it occur to you that all this was somehow related to dynamic range? Well, I really didn't want to make a big deal out of it, but just for the record... of the two of us, it wasn't me who called the other one obtuse.

[Fogel70]

I'm wondering why the 16mb K-5ii handily beats the 24mb K-3 in dynamic range by a whopping 0.7 bits in DxO's size-normalized measurements, despite having a sensor of the same size. That has nothing to do with the fact that the K-5ii has larger pixel elements? I'm not convinced.

It's basically because K.5II has lower ISO available (iSO 80) than K3. If you compare same ISO on both cameras you will see a much smaller difference.
K5II still have a small advantage over K3, but that is more likely because of other design differences on the camera/sensor than pixel size. There are many 24 MP APS-C cameras that have higher DR than K5II.

How would you explain the difference between Sony A7s and Nikon D7200, where the the APS-C sensor of 24MP give and DR advantage of 1.4EV over a FF sensor of 12MP?

[Ikarus]

How would you explain the difference between Sony A7s and Nikon D7200, where the the APS-C sensor of 24MP give and DR advantage of 1.4EV over a FF sensor of 12MP?

It looks like the A7s shines at low light/high ISO, but somehow they apparently had to give up performance at the low ISO end in order to achieve that. It seems virtually impossible to compare cameras optimized for different purposes, using sensors of different generations and manufacturers with each other based on something as basic as resolution and size. Whatever it is though, it being an FF, I don't see the A7s supporting the notion of DR just depending on the total amount of light hitting the sensor either.

[falconeye]

Well, but that's the whole point now, isn't it? The example demonstrates that you need to keep less margin for preventing clipping with the larger-area pixels, which directly translates to a higher usable dynamic range.

For the argument to hold it is not necessary to consider the FWC as the limiting factor. It still holds if the clipping happens elsewhere in the analog signal chain

If that's the whole point, then fine. Because as I said, it is a less than 0.01 EV effect.

And because FWC defines the max. analog signal by definition, there is no other clipping occuring in the signal chain.

P.S.
I've unsubscribed from the thread now.

[Rondec]

I'm not sure I understand how your example relates to the question of if and how the pixel DR affects the DR of the sensor. As far as practical examples you're asking about - I'm wondering why the 16mb K-5ii handily beats the 24mb K-3 in dynamic range by a whopping 0.7 bits in DxO's size-normalized measurements, despite having a sensor of the same size. That has nothing to do with the fact that the K-5ii has larger pixel elements? I'm not convinced.

The K5 has iso 80, while the lowest the K3 goes is iso 100. At iso 100 the K3 and K5 are the same with regard to dynamic range. The interesting thing is that the D7200 is measured as being slightly better than both of them, even though it has 24 megapixels. By your reasoning, it should not be able to do that.

[Ratcheteer]

OK then, lets look at some real world examples from Imagine Resources where you can compare a 20 MP canon 6D, with a 24 MP K-3, using similar lenses. The K-3 has a noticeably smaller pixel size, in part because it's more MP and inert because the sensor is half the surface area to begin with, so the famous double whammy.

Yet it produces distinct lines down to 2400 lw.ph. Click here for reference.

The Canon 5D mk III also tops out at about 2400 lw/ph Click here for reference.

My Pentax K-3 with a very similar lens produces 2700 lw/ph in similar conditions. Click here for reference.

Thus in using much smaller pixel size on a half size sensor. I get more resolution. How does that fit with your theory?

I have to say, these are real world examples. Not some guy wandering around in the canyons of his mind.

thats more a comparison between jpeg processing than spatial resolution

even a nikon P900 scores well above 2400 lw/ph (above 105mm mark in its zoom range) and thats with a 16mp sensor the size of 1/2.3" (6.2 x 4.6mm)

anyway this thread is so off topic, its time to let it die now.

[Ikarus]

If that's the whole point, then fine. Because as I said, it is a less than 0.01 EV effect.

I would have been curious to know how you came up with that number. I ran a few simulations, and depending on which distribution parameters I use and where I put the clipping point, I'm getting up to 0.23 for a pixel size ratio of 4, and 0.15 for a size ratio of 2, but you're right, it is small in the scheme of things.

And because FWC defines the max. analog signal by definition, there is no other clipping occuring in the signal chain.

It defines the max analog signal from the pixel. The analog signal chain includes the gain stage in the ADC. The pixel could be far from hitting the FWC and the ADC could still clip because the ISO is set too high.

P.S. I've unsubscribed from the thread now.

Thanks for hanging in there and sharing your knowledge as long as you did!