Need help with exposure question

[photo1x1.com]

Wow, I thought this thread was dead. Love the conversation (believe it or not, I read it twice ), despite it seems "a little" off topic by now.

Not being native English, I again feel the wording could be better. ISO invariance implies to me, that it would make no difference whether I push an image 1EV in Lightroom, or 5EV, which would be "100% ISO invariance". Am I wrong in this regard?

Right. ISO invariance basically means the read noise in the system no longer matters. You get the same result from digitally scaling the data in post as you would get from applying analog gain to the sensor signal prior to ADC.

When would you consider a sensor being ISO invariant? How much would you have to be able to drag the exposure slider without adding anything unwanted (noise - or what you would call it, color shift,...) to call a sensor ISO invariant?

That's why I like Bill Claff's website and test data as opposed to other sites like DXO. Do we set that standard based on what a machine can measure or on what we can practically see? Bill for example rates your camera's DR capacity at 11.6 stops. DXO rates it at 14.7 -- I believe Bill. In referring earlier to my XT-2 I said it was for all practical purposes ISO invariant. I'm sure there's a machine out there that says it's not ISO invariant. In the end all that matters to me is can I take the photo.

I have done some ISO tests a few months ago, that I repeated last week with a different background to better show the difference.
View attachment 165219

I also did an ISO invariance test with ISO400 on my Sony A7III. Reading the conversation, I should have done it at ISO100, or ISO 640 (according to this graph Photographic Dynamic Range versus ISO Setting), is that correct?

Correct. Your camera is most certainly not ISO invariant and very ISO invariant both at the same time -- just like my Fuji. Your Sony is fitted with a dual impedance sensor. The sensor has two read channels. You've got ISO invariance in each read channel separately but not across them. Your test here compares the two channels one to the other -- NOT ISO invariant.

Oh yes, and one more question, if somebody would be willing to answer: why is the input-referred read noise in this graph: photonstophotos.net/Charts/RN_e.htm#Sony ILCE-7M3_14 higher for lower ISOs?

Dual impedance sensor -- it's really like having two sensors in the camera, one for low ISOs and another one for high ISOs.

For those of you interested, I have uploaded the RAW files here: http://www.amriphoto.com/forum/ISOInvarianceTest.zip
In regard to noise, I feel the ISO400 files keep up pretty well up to pushing them +2EV in Lightroom and comparing them to ISO3200. But to be honest, I do see quite a difference starting at +3EV when pixel peeping.
So my conclusion is no real surprise: you have to decide when shooting: is it more important for me to have a higher dynamic range or less noise. In the particular case of my orchid, I would have gone with less noise.

I know this is not a scientific test as it was done with (quality though) LED lights, but it should give a good impression.
Here is a sample - which one do you think is ISO400 +3EV in Lightroom, and which is ISO3200?

Sorry to do this to you but you need to re-do the test and stay in one or the other of the two impedance channels on the sensor.

Two more concerns with your test data: 1. If you look at all the EV 0 exposures in RawDigger you're nearly a full stop underexposed (defined in this case as reaching sensor saturation). 2. Sony is notorious for cooking their raw files and slipping in some lossy compression. I don't know specifically about your camera but that may be a factor.

There's another complication to this topic that needs to be considered and that's the external processing software. All raw converters don't take the same approach to what they're doing and it can matter a lot in this regard. How to assign value to this difference is personal so I think both choices are entirely valid. If you've ever spent time programming a computer you know that there can be a huge difference in performance based on math precision. If you really want to keep computer software moving fast make sure all the math stays integer based. Nothing slows down a computer processor like math with floating point numbers.

The above said processing speed is a critical issue for raw conversion software and I believe LR has its thumb on the scale tipping toward speed over precision. You may be quite surprised to compare the same test processed in LR and then again in say C1 or SilkyPix.

Joe

Thanks, Joe - much appreciated! I redid the test right away. Luckily my orchid wasn´t dead yet.
I exposed it one stop brighter, and used ISO100 and ISO640 for the test. I also checked the images in capture one. Still, starting at +3EV noise starts being pretty visible compared to the higher ISO shot. I didn´t use capture one in a while and was impressed by their noise removal algorithm.
For those interested, here are the RAW-files: www.amriphoto.com/forum/ISOInvarianceTest2.zip

May I refine my question:

Oh yes, and one more question, if somebody would be willing to answer: why is the input-referred read noise in this graph: photonstophotos.net/Charts/RN_e.htm#Sony ILCE-7M3_14 higher for lower ISOs?

Dual impedance sensor -- it's really like having two sensors in the camera, one for low ISOs and another one for high ISOs.

I did know about the dual impendance, having read quite a bit about ISO invariance recently. However, I wonder why the input-referred read noise is higher in lower ISO-regions, I would have expected it the other way around. That confuses me. What exactly is that input-referred read noise?

 

[Ysarex]

It's a shame that the truth sometimes takes so many words to explain. Joe's two excellent posts above show why the so-called exposure triangle is no longer applicable, now that ISO invariance completely invalidates one of the critical parameters of the so-called exposure triangle.

Isn't this expose for the highlights just like in chrome films?

Yes.

YOu then adjust the sahdow using the shadow slider just as before. The only difference is the better sensors today allow better results. If you expose for the shadows, the highlights clip. Same with digital same with chromes. Raising or lowering the ISO has no effect on the large difference in stops between shadows and highlights.

Raising or lowering the ISO has no effect on the difference in stops in the scene being photographed. It does have an effect on the range of stops that film or a digital sensor can record. With film moving to a higher ISO film INCREASES the range of stops you can record. With digital raising the ISO DECREASES the range of stops you can record.

That's why we use graduated ND filters or DHR. If the sensor allows a greater range because of it's DR, that's just another advantage of a better sensor. But settings still have to be adjusted based on a triangle of settings. I don't see what's changed. Maybe I'm missing something.

The film/sensor difference relative to ISO involves light sensitivity. Different films with different ISO values really are more or less light sensitive. Using film, especially chrome, it's critical to not over or under expose it. Changing ISO with a digital sensor does not alter the light sensitivity of the sensor and if the sensor is ISO invariant then it doesn't matter at all if you do this:

1/100 sec f/8 ISO 1600

or this:

1/100 sec f/8 ISO 100

as long as the ISO 1600 shot doesn't clip the diffuse highlights.

If on the other hand you do have a high contrast scene with a range of let's say 9 stops and you're using Wolfgang's Sony camera from a couple posts up you can record all 9 stops of that scene with the camera set to ISO 100 but you can't record those nine stops if you raise the ISO to 1600.

Joe

 

[photo1x1.com]

When would you consider a sensor being ISO invariant? How much would you have to be able to drag the exposure slider without adding anything unwanted (noise - or what you would call it, color shift,...) to call a sensor ISO invariant?

That's why I like Bill Claff's website and test data as opposed to other sites like DXO. Do we set that standard based on what a machine can measure or on what we can practically see? Bill for example rates your camera's DR capacity at 11.6 stops. DXO rates it at 14.7 -- I believe Bill. In referring earlier to my XT-2 I said it was for all practical purposes ISO invariant. I'm sure there's a machine out there that says it's not ISO invariant. In the end all that matters to me is can I take the photo.

BTW: these words surprise me quite a bit

 

[photo1x1.com]

@american
Thanks, Tim! I do understand ISO invariance (well, at least I think so ). But ISO invariance would mean, that there is absolutely no noise produced in camera after the analog amplification process. So it wouldn´t matter whether you let the camera do the digital amplification, or the software. But that´s the ideal case which is not the reality in current cameras it seems. Still people talk about ISO invariance. I´d call it relatively invariant, or you´d have to find some kind of index, like LEDs have cri (Color Rendering Index). Because if you call the current sensors ISO invariant, how would you call the ones in 10-20 years "proper invariant" ? So that´s probably the next chance to create misunderstandings. Just like "exposure", "camera exposure", "proper exposure",...

An ISO Invariant sensor has no analog gain. It does not mean no noise.
And yes, we are currently in a transition. Some companies have switched to hybrid ISO with some analog and some digital. Such as when you read the sensor primer, the specific Nixon is a hybrid. I know my Canon 6D is also.
Per @Ysarex there are a few camera's which are sensor invariant, and others which are relatively close. He named a couple of them, but I need to run into a meeting so I cannot go back and dig them out of the thread.

This transition period is actually the "saving" grace for my wallet right now, and what stopped me from ordering the Canon EOS R so I can keep my same high quality native lenses. Since it looks like I am going to switch systems, I may as well wait....

Tim

Tim, the analog gain remains no matter if a sensor is ISO invariant or not (if I´m not totally wrong), because it is done before analog is converted to digital (which is essential in digital photography).

 

[Ysarex]

Wow, I thought this thread was dead. Love the conversation (believe it or not, I read it twice ), despite it seems "a little" off topic by now.

Not being native English, I again feel the wording could be better. ISO invariance implies to me, that it would make no difference whether I push an image 1EV in Lightroom, or 5EV, which would be "100% ISO invariance". Am I wrong in this regard?

Right. ISO invariance basically means the read noise in the system no longer matters. You get the same result from digitally scaling the data in post as you would get from applying analog gain to the sensor signal prior to ADC.

When would you consider a sensor being ISO invariant? How much would you have to be able to drag the exposure slider without adding anything unwanted (noise - or what you would call it, color shift,...) to call a sensor ISO invariant?

That's why I like Bill Claff's website and test data as opposed to other sites like DXO. Do we set that standard based on what a machine can measure or on what we can practically see? Bill for example rates your camera's DR capacity at 11.6 stops. DXO rates it at 14.7 -- I believe Bill. In referring earlier to my XT-2 I said it was for all practical purposes ISO invariant. I'm sure there's a machine out there that says it's not ISO invariant. In the end all that matters to me is can I take the photo.

I have done some ISO tests a few months ago, that I repeated last week with a different background to better show the difference.
View attachment 165219

I also did an ISO invariance test with ISO400 on my Sony A7III. Reading the conversation, I should have done it at ISO100, or ISO 640 (according to this graph Photographic Dynamic Range versus ISO Setting), is that correct?

Correct. Your camera is most certainly not ISO invariant and very ISO invariant both at the same time -- just like my Fuji. Your Sony is fitted with a dual impedance sensor. The sensor has two read channels. You've got ISO invariance in each read channel separately but not across them. Your test here compares the two channels one to the other -- NOT ISO invariant.

Oh yes, and one more question, if somebody would be willing to answer: why is the input-referred read noise in this graph: photonstophotos.net/Charts/RN_e.htm#Sony ILCE-7M3_14 higher for lower ISOs?

Dual impedance sensor -- it's really like having two sensors in the camera, one for low ISOs and another one for high ISOs.

For those of you interested, I have uploaded the RAW files here: http://www.amriphoto.com/forum/ISOInvarianceTest.zip
In regard to noise, I feel the ISO400 files keep up pretty well up to pushing them +2EV in Lightroom and comparing them to ISO3200. But to be honest, I do see quite a difference starting at +3EV when pixel peeping.
So my conclusion is no real surprise: you have to decide when shooting: is it more important for me to have a higher dynamic range or less noise. In the particular case of my orchid, I would have gone with less noise.

I know this is not a scientific test as it was done with (quality though) LED lights, but it should give a good impression.
Here is a sample - which one do you think is ISO400 +3EV in Lightroom, and which is ISO3200?

Sorry to do this to you but you need to re-do the test and stay in one or the other of the two impedance channels on the sensor.

Two more concerns with your test data: 1. If you look at all the EV 0 exposures in RawDigger you're nearly a full stop underexposed (defined in this case as reaching sensor saturation). 2. Sony is notorious for cooking their raw files and slipping in some lossy compression. I don't know specifically about your camera but that may be a factor.

There's another complication to this topic that needs to be considered and that's the external processing software. All raw converters don't take the same approach to what they're doing and it can matter a lot in this regard. How to assign value to this difference is personal so I think both choices are entirely valid. If you've ever spent time programming a computer you know that there can be a huge difference in performance based on math precision. If you really want to keep computer software moving fast make sure all the math stays integer based. Nothing slows down a computer processor like math with floating point numbers.

The above said processing speed is a critical issue for raw conversion software and I believe LR has its thumb on the scale tipping toward speed over precision. You may be quite surprised to compare the same test processed in LR and then again in say C1 or SilkyPix.

Joe

Thanks, Joe - much appreciated! I redid the test right away. Luckily my orchid wasn´t dead yet.
I exposed it one stop brighter, and used ISO100 and ISO640 for the test. I also checked the images in capture one. Still, starting at +3EV noise starts being pretty visible compared to the higher ISO shot.

Could be a difference in Sony's cooked raw files?

I didn´t use capture one in a while and was impressed by their noise removal algorithm.
For those interested, here are the RAW-files: www.amriphoto.com/forum/ISOInvarianceTest2.zip

May I refine my question:

Oh yes, and one more question, if somebody would be willing to answer: why is the input-referred read noise in this graph: photonstophotos.net/Charts/RN_e.htm#Sony ILCE-7M3_14 higher for lower ISOs?

Dual impedance sensor -- it's really like having two sensors in the camera, one for low ISOs and another one for high ISOs.

I did know about the dual impendance, having read quite a bit about ISO invariance recently. However, I wonder why the input-referred read noise is higher in lower ISO-regions, I would have expected it the other way around. That confuses me. What exactly is that input-referred read noise?

Time to ask Bill -- I know why the difference is there but I'm not able to explain further.

Joe

 

[photo1x1.com]

Thanks, Joe - I appreciate the time you have put into this matter!

 

[Ysarex]

When would you consider a sensor being ISO invariant? How much would you have to be able to drag the exposure slider without adding anything unwanted (noise - or what you would call it, color shift,...) to call a sensor ISO invariant?

That's why I like Bill Claff's website and test data as opposed to other sites like DXO. Do we set that standard based on what a machine can measure or on what we can practically see? Bill for example rates your camera's DR capacity at 11.6 stops. DXO rates it at 14.7 -- I believe Bill. In referring earlier to my XT-2 I said it was for all practical purposes ISO invariant. I'm sure there's a machine out there that says it's not ISO invariant. In the end all that matters to me is can I take the photo.

BTW: these words surprise me quite a bit

A couple times in this thread I stuck in a paragraph and used the word perspective. This makes me think of Derrel who posted a couple times in this thread. He does a good job of keeping perspective. It's too easy to nerd out on this stuff to the detriment of getting the photo. Always place getting the photo as the highest priority.

Joe

 

[Ysarex]

If an a variance sensor allows you to keep the iso at 100, let's say, you still work with triangle. It's like always using 100 ASA (ISO) film.

It's like always using ISO 100, 200, 400, 800, 1600, 3200 film all at once in every photo. The point of ISO invariance is you can ignore ISO entirely and that means you can ignore the light meter settings as long as you're within the rough range of those 5 stops between 100 and 3200.

You raise ISO on a digital camera when you can't get the other two settings you want -- typically a fast enough shutter speed. If you raise the shutter speed you're going to expose less. ISO invariance means you can go ahead and do that, raise the shutter speed to where you need it and ignore changing the ISO because it doesn't matter. Just change the shutter speed and keep shooting. You'll expose less but there's no need to bother with the ISO because you're going to get the same photo if you leave it at 100 or raise it to 800. Just take the photo since the ISO doesn't matter.

Joe

You still have to set the light meter's ASA(ISO) to 100 to get the other two settings of speed and aperture.

 

[photo1x1.com]

Time to ask Bill -- I know why the difference is there but I'm not able to explain further.

Joe

Wow, Bill replied really quick - here is his answer:

Input-referred Read Noise always drops as you increase amplification.
This is true of all cameras.



The big drop at ISO 640 is where High Conversion Gain (HCG) on this dual conversion gain sensor begins.
The drop above ISO 51200 is due to Noise Reduction (NR).

 

[Ysarex]

Time to ask Bill -- I know why the difference is there but I'm not able to explain further.

Joe

Wow, Bill replied really quick - here is his answer:

Input-referred Read Noise always drops as you increase amplification.
This is true of all cameras.



The big drop at ISO 640 is where High Conversion Gain (HCG) on this dual conversion gain sensor begins.
The drop above ISO 51200 is due to Noise Reduction (NR).

Bill is awesome! We should all stop by periodically and leave a thank you note.

Joe

 

[n614cd]

Tim, the analog gain remains no matter if a sensor is ISO invariant or not (if I´m not totally wrong), because it is done before analog is converted to digital (which is essential in digital photography).

A truly ISO invariant sensor would have no variable analog gain. Based on what I have seen, such a sensor does not yet exist.
However, there are more sensors which are getting closer and only have limited number of steps.

Reminder, I am new to this level of detail so I am doing discovery! So do not take what I state as gospel!

Tim

 

[AlanKlein]

The issue about using the triangle to determine exposure came up before invariant sensors. Since 99.9% if cameras are not invariant, then you still need to enter the appropriate ISO. If not, what am I missing?

 

[Overread]

Even with an ISO Invariant sensor most photographers will still err toward proper exposure in camera. It provides the best way to review the shot on the back and honestly its FAR more pleasing to sit down and tweak an already good shot than it is to have to adjust everything on a black disaster. Sure the invariance gives you FAR more leeway, esp in difficult or very shifting lighting conditions; but I'd wager even with magical sensors we'd still mostly aim for correct in camera exposure.

 

[Ysarex]

Even with an ISO Invariant sensor most photographers will still err toward proper exposure in camera. It provides the best way to review the shot on the back and honestly its FAR more pleasing to sit down and tweak an already good shot than it is to have to adjust everything on a black disaster. Sure the invariance gives you FAR more leeway, esp in difficult or very shifting lighting conditions; but I'd wager even with magical sensors we'd still mostly aim for correct in camera exposure.

Of course. Nowhere in this thread has anyone made a suggestion otherwise. I took a photo yesterday and posted it here earlier this afternoon. :White on Red I set the ISO before I took the photo for the very reasons you note.

Joe

 

[Ysarex]

The issue about using the triangle to determine exposure came up before invariant sensors. Since 99.9% if cameras are not invariant,

That depends on how you're going to define ISO invariance. You can have a machine measure it and get your figure above or you can measure it visually like this guy did and get a pretty good list of cameras: ISO Invariance: What it is, and which cameras are ISO-less

Fortunately you have an opportunity in this thread to back up your 99.9% figure by going to post #61 in this thread: Need help with exposure question and explaining what to look at in those two photos that makes it obvious why the ISO 100 photo (underexposed 3 stops) is worse than the ISO 800 photo that was "properly exposed." If you can't do that then your 99.9% figure is bogus as far as real world usage is concerned.

then you still need to enter the appropriate ISO. If not, what am I missing?

Can't imagine. Something about still not understanding why the Exposure Triangle is wrong and confuses people?

Joe