I have been uploading files to Bill Claff (@bclaff) so he can add the Sony A1II to his Photonstophotos web site. He took a look at the initial data and informed me that the ISO for the second gain is 640, which is a change from ISO 500 in the A1. I am surprised since the claim is that the sensor is unchanged. It is only firmware that sets the ISO at which the sense node capacitance is changed, but the actual capacitance of the sense node is determined by the sensor design. If the capacitance is unchanged it is not clear to me how the noise performance of the sensor is improved at higher ISO. There have been claims that the high ISO performance is somewhat improved so the increase from 500 to 640 is consistent with that, but seems inconsistent with the claim the sensor is unchanged. I assume we will learn more as we get Bill's analysis, and get more details about the sensor.
Regardless, I prefer that the change is at a higher ISO since I operate essentially all the time at or above 500 in the A1 and prefer that the second gain be a bit higher. It looks like the A1II is an improvement for my style of shooting and I believe many others are the same.
dclark wrote:
The Photonstophotos site has been updated by @bclaff@ with the A1II data. Input_referred_read_noise Photographic_Dynamic_Range
The A1II is slightly better at all ISO's (except for ISO 500 due to the change in dual gain).
Interesting!
Maybe not very interesting in practical terms (to me) but as interested in the technical matters in general I get curious. I have neither tried to measure noise , nor have I been interested in noise and I very seldom use any kind of denoising. This to explain I know nothing about how it works or what the photons to photos curves are based on. But I have a question or two.
What a great site! I played around with curves, cameramodels and different ways to show noise numbers in different ways. I read parts of the primers.
(1) I don't get the ISO values. In the primers the ISO values are the same as those we set our cameras to. In the current diagrams the ISO values differ a bit up and down from the standard values. Is the "true" ISO value somehow calculated or estimated in the diagrams? I guess they are as we otherwise would compare apples to oranges all the time.
(2) Looking at the diagrams and the DPR Image quality thing checking the R5ii I'm under the impression there is some noise cancelling going on in the raw files. Is that correct?
Has anyone had time to experiment with the best settings for higher ISO shots?
Should the camera noise reduction be set to low / standard / off?
Should the various lens corrections be sent to auto / off / ...?
(ie, does the camera make it easier or harder to post-process).
Clearly, if you want just out of camera JPEGs then in-camera processing is the way to go.
I only ask because I see that at higher ISOs, if I test the Lightroom sharpening mask view, there are obvious grid-like patterns in the darker areas, some of these presumably the result of in-camera processing.
Jonas B wrote:
Interesting!
Maybe not very interesting in practical terms (to me) but as interested in the technical matters in general I get curious. I have neither tried to measure noise , nor have I been interested in noise and I very seldom use any kind of denoising. This to explain I know nothing about how it works or what the photons to photos curves are based on. But I have a question or two.
What a great site! I played around with curves, cameramodels and different ways to show noise numbers in different ways. I read parts of the primers.
(1) I don't get the ISO values. In the primers the ISO values are the same as those we set our cameras to. In the current diagrams the ISO values differ a bit up and down from the standard values. Is the "true" ISO value somehow calculated or estimated in the diagrams? I guess they are as we otherwise would compare apples to oranges all the time.
(2) Looking at the diagrams and the DPR Image quality thing checking the R5ii I'm under the impression there is some noise cancelling going on in the raw files. Is that correct? ...Show more →
1) Those are the real ISO values: all the (power 2) x 100 ISOs are exact (200, 400, 800, 1600, 3200 ... 51200) because they're full stops. The ones that are not exact are not (powers of 2) x 100, because they're either a third of a stop, or 2 thirds, and due to Math ... Pow(2, 2 + 1/3) * 100 => 503.9, and not 500 => 503 is correct, and 500 is shown in camera but actually it's really 503. Same for 251.9, etc.
dclark wrote:
The Photonstophotos site has been updated by @bclaff@ with the A1II data. Input_referred_read_noise Photographic_Dynamic_Range
The A1II is slightly better at all ISO's (except for ISO 500 due to the change in dual gain).
Do you think the dynamic range is better because the read noise is better? I’m guessing the reduced read noise is due to reduced noise in the circuits near the sensor. So it could be the same sensor with improvement elsewhere?
octo wrote: 1) Those are the real ISO values: all the (power 2) x 100 ISOs are exact (200, 400, 800, 1600, 3200 ... 51200) because they're full stops. The ones that are not exact are not (powers of 2) x 100, because they're either a third of a stop, or 2 thirds, and due to Math ... Pow(2, 2 + 1/3) * 100 => 503.9, and not 500 => 503 is correct, and 500 is shown in camera but actually it's really 503. Same for 251.9, etc.
Ah, I see. I should have figured that out.
So, this is still only values taken from the camera settings then? Not something measured, kind of true value or exposed and developed the same way? Of course, I don't know if it technically would be possible to compare that way... What I'm trying to say is that if the ISO values in the graphs just are what respectively camera says it is I would find it hard to believe in them.
Jonas B wrote:
Ah, OK. But, however it is done, applying some kind of noise reduction to raw files seems as a stupid idea.
It's always been that way though, and I wouldn't just dismiss it as a "stupid idea." Clearly there are reasons all of the camera manufacturers choose to do so at some point on the ISO scale, and they know vastly more about their sensors, etc. than we do.
jhapeman wrote:
It's always been that way though, and I wouldn't just dismiss it as a "stupid idea." Clearly there are reasons all of the camera manufacturers choose to do so at some point on the ISO scale, and they know vastly more about their sensors, etc. than we do.
Then that just goes to show how little I know about the technical sensor details. Perhaps the stupid (I really mean it) anti vignetting and star-eating habit have made me sensitive when it comes to manipulated raw files.
Perhaps everything isn't stupid. Then again, I see camera makers do really stupid things so I'm not sure everything they do is with the users best in mind.
Why is it better to reduce noise in camera instead of in post? (Somewhat rethorical, no need to reply.)
Dec 25, 2024 at 02:01 PM
Steve Spencer Offline • • • • • • • Upload & Sell: On
octo wrote:
Ah, I see. I should have figured that out.
So, this is still only values taken from the camera settings then? Not something measured, kind of true value or exposed and developed the same way? Of course, I don't know if it technically would be possible to compare that way... What I'm trying to say is that if the ISO values in the graphs just are what respectively camera says it is I would find it hard to believe in them.
Ah, OK. But, however it is done, applying some kind of noise reduction to raw files seems as a stupid idea. ...Show more →
On the site somewhere Bill Claff, discusses why he uses the camera's reported ISO rather than measured ISO, but I can't find it right now. I could find this analysis in which he did a study of DXO's measured ISOs vs. camera settings for ISO here:
And basically he finds most cameras are off by about a third of a stop and he believes that is to protect highlights and not to game the system by falsifying ISO. Basically, I think he is suggesting that the camera values are a good place to start and that measuring ISO wouldn't change things much, but I may be distorting his argument.
tschopp wrote:
Do you think the dynamic range is better because the read noise is better? I’m guessing the reduced read noise is due to reduced noise in the circuits near the sensor. So it could be the same sensor with improvement elsewhere?
Hard to say. If the noise contributions are downstream of the sense node then the the input-referred-read-noise will have a downward slope as ISO increases. The A1 and the A1II are essentially level which means Sony has done a good job eliminating downstream noise. The read noise is so low (~4 photoelectrons, and ~1 photoelectron) it is hard to see how they can make much improvement. The slight improvement in the A1II is both below and above the dual gain transition, so that suggests it is not just due to a change in sense node capacitance. We lack detailed information so anything we may say is speculation. It is clear that the noise performance we are getting with both the A1 and A1II is close to perfect and there is not much room for improvement.
@bclaff asked that I send him electronic shutter, ES, files to supplement the mechanical shutter files he had already analysed. This was to see if there was any difference in read noise. He just sent me a note saying the ES read noise is slightly higher, but the surprise was that the dual gain transition returned to ISO 500. I think he will publish that data on his site soon.
Steve Spencer wrote:
[...] Basically, I think he is suggesting that the camera values are a good place to start and that measuring ISO wouldn't change things much, but I may be distorting his argument.
I don’t have the knowledge to comment on why mechanical and electronic shutter should behave differently in this respect. What I know for sure is that I had almost forgotten about the mechanical shutter and I don’t need reasons to be tempted to go back to it 😌.
Let’s see if there is any real life difference in noise and DR between mechanical and electronic shutters. Dpreview studio scene samples are always a useful addition to bclaff’s tests.
dclark wrote:
@bclaff@ asked that I send him electronic shutter, ES, files to supplement the mechanical shutter files he had already analysed. This was to see if there was any difference in read noise. He just sent me a note saying the ES read noise is slightly higher, but the surprise was that the dual gain transition returned to ISO 500. I think he will publish that data on his site soon.
dclark wrote:
I have been uploading files to Bill Claff (@bclaff) so he can add the Sony A1II to his Photonstophotos web site. He took a look at the initial data and informed me that the ISO for the second gain is 640, which is a change from ISO 500 in the A1. I am surprised since the claim is that the sensor is unchanged. It is only firmware that sets the ISO at which the sense node capacitance is changed, but the actual capacitance of the sense node is determined by the sensor design. If the capacitance is unchanged it is not clear to me how the noise performance of the sensor is improved at higher ISO. There have been claims that the high ISO performance is somewhat improved so the increase from 500 to 640 is consistent with that, but seems inconsistent with the claim the sensor is unchanged. I assume we will learn more as we get Bill's analysis, and get more details about the sensor.
Regardless, I prefer that the change is at a higher ISO since I operate essentially all the time at or above 500 in the A1 and prefer that the second gain be a bit higher. It looks like the A1II is an improvement for my style of shooting and I believe many others are the same....Show more →
I think your way off as to how a dual gain sensor works. you just turn the op amps on a little harder, standard iso is controled after the ADC not at sensor level.
Dec 25, 2024 at 03:37 PM
Steve Spencer Offline • • • • • • • Upload & Sell: On
A74me wrote:
I think your way off as to how a dual gain sensor works. you just turn the op amps on a little harder, standard iso is controled after the ADC not at sensor level.
No, dclark has it right. Dual gain occurs at the sensor level. See this white paper for the details:
