Sony high ISO performance

These examples were produced by Pierre Sottas from D7000 files. I provided the link to his post upthread, but here it is again: http://forums.dpreview.com/forums/readflat.asp?forum=1034&thread=36887819

I have not copied the images but inserted links to their location (and most browsers will then display the image instead of the link)

I have also linked to my own humble test from the a900 previously in this thread.
Here are some from an earlier test that I did myself:

http://forums.dpreview.com/forums/read.asp?forum=1037&message=38125800
 
I will never again say that ACR is useless for this exercise although I have not been able to get such close results using ACR myself.
 
What does "consensus" mean? I'd like to see a link to real proof of this, not repeated internet theories. EV compensation is made by changing shutter speed and/or aperture, not by changing iso setting, with the possible exception of when in auto iso mode.
Among people who post on these forums who seem to know more than I do. That's why I said the consensus "seems to be" as I am not absolutely sure. I would like to hear from a engineer who actually designs camera circuits to know for sure but, since I do have a degree in Electrical Engineering Technology from some time ago, I tend to believe that the gain occurs after the signal is sent from the sensor and the amount of gain applied is relative to the amount of light received by the sensor. The amount of current produced by the sensor must be extremely small so there is, I believe, gain applied by analog circuits to allow the camera to have a strong enough signal for digital processing or the signal would be buried in the "noise". This amplification must be applied before A/D conversion (sensors produce analog signals that must be converted to digital). One of the advantages of CMOS sensors is the necessary circuitry is built on the chip and therefore provides the shortest possible analog circuit path. It is within that circuit path that noise is introduced and the longer the path the greater chance for noise. The more gain that is applied the more that noise will be amplified as well. The greater the amount of light on the sensor the greater the signal strength produced by the sensor which will then require less amplification and therefore there will be a greater S/N ratio. In the end because 12800 iso at +2 EV allows the same amount of light to excite the sensor as iso3200 at 0EV as far as the sensor signal is concerned there is no difference between the two. BTW, you seem to agree with me so why are you balking at what I said?

--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
Iif you apply +2 EV you get more light on the sensor because you overexpose (relative to the 'normal' 0 EV exposure). Try comparing a normally metered 3200 shot with a 12800 one with +2 EV and you will see what I mean even though the aperture and shutter were the same.
The two shots will be exactly the same in a RAW shot. There may be a difference in the way the signal is processed for jpg however since the camera digital processing may be different at higher iso's. If that's the case you are better off using iso3200 with 0EV comp because more processing at higher iso is more destructive.

--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
No, we just hate to be right when getting rid of a some common misunderstandings (involving sensors). ;)
So you think the light arriving at the sensor is digitized first and then amplified via software? I would like to see the proof of this as all I see is people quoting what they read in a forum.
I am no expert on the workings of sensors but if this was true it would make for distortion as at low light levels you would only be using a few bits of the ADC. I can only relate to my experience with sound recording where you need to preamp the signal to an optimum level (iso setting) to use all the bits of the ADC and get the best sound (picture).
If you think of the sensor as a microphone the theory is exactly the same. The sensor is an analog device. CMOS sensors have the circuitry built on the chip to boost the analog signal and, because the A/D circuits are built into the chip, to then convert to digital. There digital amplification can be applied after A/D conversion.
--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
I will never again say that ACR is useless for this exercise although I have not been able to get such close results using ACR myself.
These are processed pretty quick. With some more post processing the differences would be even less visible!
 
In the end because 12800 iso at +2 EV allows the same amount of light to excite the sensor as iso3200 at 0EV as far as the sensor signal is concerned there is no difference between the two. BTW, you seem to agree with me so why are you balking at what I said?
Because EV compensation is applied with shutter or aperture changes, not iso gain.
There is more light, not more iso gain.

Cheers,
--

"Those who would sacrifice liberty for safety deserve neither liberty nor safety." - Ben Franklin.
You can see larger versions of my pictures at http://www.dennismullen.com .
 
... There digital amplification can be applied after A/D conversion.
But is it?
As I said, the iso gain is probably applied before A/D conversion.
As I said previously the current generated by the actual sensor is very small so yes there must be a signal gain before A/D conversion but whether this gain varies with iso is something I don't know. It has been claimed by others that the 16mp Sony chip is iso100 only and all gain is digital but if that is true I don't know.

--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
In the end because 12800 iso at +2 EV allows the same amount of light to excite the sensor as iso3200 at 0EV as far as the sensor signal is concerned there is no difference between the two. BTW, you seem to agree with me so why are you balking at what I said?
Because EV compensation is applied with shutter or aperture changes, not iso gain.
There is more light, not more iso gain.
Yes that is true and I thought that's what I was saying. With more light the sensor produces more current so less gain is needed irregardless of iso.

--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
As I said previously the current generated by the actual sensor is very small so yes there must be a signal gain before A/D conversion but whether this gain varies with iso is something I don't know. It has been claimed by others that the 16mp Sony chip is iso100 only and all gain is digital but if that is true I don't know.
It's probably a good idea not to state it as fact then.

Cheers,
--

"Those who would sacrifice liberty for safety deserve neither liberty nor safety." - Ben Franklin.
You can see larger versions of my pictures at http://www.dennismullen.com .
 
The gain is set by the choice of ISO setting. By then exposing at +2EV you are overexposing in most circumstances. Using RAW you may be able to recover the highlights and end up with a less noisy picture. The is what the OP was saying, and I completely agree with him. If you don't believe, try it

cheers

tom
 
Iif you apply +2 EV you get more light on the sensor because you overexpose (relative to the 'normal' 0 EV exposure). Try comparing a normally metered 3200 shot with a 12800 one with +2 EV and you will see what I mean even though the aperture and shutter were the same.
Assuming the metering is correct this will clip the highlights.
Except that if you are using RAW there is always some headroom for highlight recovery, actually usually a surprising amount providing you don't overdo it. Also there is much more data captured at the right hand end of the histogram (there was an article about this a few weeks ago) so by exposing as far as you can to the right without going beyond the point where highlights can be recovered, you have more data to manipulate in post.
It's a good discussion though, because it leads to the point that you need to expose as much to the right as you can (and still recover needed highlights in PP) if you are looking for low noise.
I have found that I get better color fidelity, less clipped highlights, and less work being more conservative with my exposure, especially in high contrast situations.
Again it depends on your PP technique of RAW images
In low light situations where noise is a problem there is no leeway. If you could boost the exposure you could also lower the ISO setting.
In my experience it is better to get the shadows more exposed as that is where the noise lies. as said above you can tame the highlights later (providing you don't overdo it)
The ability to recover highlights is very limited in JPG, and that is why the Agorabasta JPG settings were so popular, effectively they reduced the highlights in JPG, allowing more exposure without blowing (so far as I can analyse their effect).
I never could understand doing this as using Raw would be a better solution.
Absolutely agree
When comparing the high ISO/low noise capabilities of different cameras, this is something which needs normalising. If camera A normally overexposes relative to camera B it is going to appear as though it is a better low light noise performer.
Proper exposure is important and not camera specific.
True, but consider that LR boosts brightness by 50 in its defaults for Sony images, presumably because Adobe considers them to be too dark. If they were a bit overexposed and brightness was reduced they would look better....you can try it to see what I mean. If another make of camera exposes more, their noise characteristics are going to look better in superficial comparisons.
Cheers as well...good discussion!

Tom
--

"Those who would sacrifice liberty for safety deserve neither liberty nor safety." - Ben Franklin.
You can see larger versions of my pictures at http://www.dennismullen.com .
 
It's probably a good idea not to state it as fact then.
Now you.re just being a dickhead. I stated I am not sure about any of this in more that one post since I am not an expert on camera electronics . Please show me where I stated what I said is fact. I do believe however that what I did say, based on what I have read, is mostly true. I still know though that iso3200 at 0EV is the same as iso12800 at +2 for a RAW photo due to the same amount of light hitting the sensor.
--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
The gain is set by the choice of ISO setting. By then exposing at +2EV you are overexposing in most circumstances. Using RAW you may be able to recover the highlights and end up with a less noisy picture. The is what the OP was saying, and I completely agree with him. If you don't believe, try it
I did try it and at iso12800 +2 I got the exact same shutter speed and f number as iso3200 at 0EV so the 2 shots received the same amount of light. In addition the two photos looked exactly the same even at 100%.
--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
WOW! So many interesting comments and opinions. I appreciate you guys taking interest in this post. I am learning a lot and at the same time have more questions rise up. There are a few people here that seem very certain that ISO 12800 +2EV with the same shutter and f/stop is equivalent to that of 3200 0EV. Well I'm not going to agree or disagree just yet. I didn't have much time last night but I picked up my camera and snapped two pictures of the same stuffed toy dog. The first was shot at ISO 3200, shutter 1/160, f/stop 2.8 and 0EV. The second of course ISO 12800, shutter 1/160, f/stop 2.8 and +2EV and I was not surprised to see the results. After that I opened the first image in Sony's RAW converter that came with the camera and I hardly use and saved the RAW file as a JPEG. Then I opened the second image in LR3 which I normally use for exposure adjustments and brought the exposure down to match the first image. Then saved it as a JPEG. Then I was surprised! The two images where quite different as far as noise goes. The ISO 3200 image had a lot more chroma or color noise. The ISO 12800 was a little grainy but practically had no color noise (interesting). Also I niticed more detail in the second shot. This might have a lot to do with the software. I will not jump to any conclusions at this time until I do some further testing. I should have some time this weekend to make some comparison shots and then I will post them and let you guys decide for yourselves. In the mean time if any of you guys have some time please do some of your own testing and Post! Post! Post!
 
After that I opened the first image in Sony's RAW converter that came with the camera and I hardly use and saved the RAW file as a JPEG. Then I opened the second image in LR3 which I normally use for exposure adjustments and brought the exposure down to match the first image. Then saved it as a JPEG. Then I was surprised! The two images where quite different as far as noise goes.
To get a proper comparison you must use the same RAW converter. Different RAW converters perform very differently from one another. Also all image processing should be set to off or minimum for a proper comparison or all you are seeing is results from post processing. I did that with a test just like you did using PSPX3 and minimum processing and the two photos were virtually indistinguishable.

--
Tom

Look at the picture, not the pixels

http://www.flickr.com/photos/63683676@N07/
http://www.flickr.com/photos/25301400@N00/
 
I really can't understand that, I have just tried it again myself (simulated see below). Of course the aperture/shutter speed were the same at both ISO settings as expected (the +2EV increases the exposure time to be the same as at 3200) but the shot taken at 12800 is greatly overexposed .

As the op says in a later recent post if you then correct the overexposure in LR you end up with a reasonable shot which may be better than the 3200 one depending on circumstances. As I said earlier, there is much more data at the right hand end of the histogram, and detail can indeed be improved by exposing to the right

(Note that as I use an a700 I have done the test using 1600 and 6400 ISO, but this does not make any difference to the results)

I do suggest you try again and make sure the +2 is properly applied as what you are reporting must indicate a fault in your camera

Thanks for trying, anyway

tom
 

Keyboard shortcuts

Back
Top