Iso rating and post - not exactly like a DSLR

[Fredrik Harreschou]

Hello to you all.

This is my first post here, even though I've been surfing this forum everyday for the last six months. Great forum. I'm picking up something new and interesting every day.

Here's my question:
In this thread http://www.reduser.net/forum/showthread.php?t=5181&page=2 once again the confusion was evident concerning shooting vs processing at higher ISOs. Graeme responded:


Just to be very clear, he displayed and probably processed the image at ASA 1000. He did not record it at ASA 1000 since we don't change the data coming from the sensor.

The ISO/ASA change is all done after the data has been captured. It's a digital 'gain' (exposure change).

I just realised this is different from a DSLR. Shooting at a higher ISO will give you a much better (cleaner) image than adding "digital gain"/exposure when converting the raw file. Here's an example:

This is shot at ISO1600 and slightly underexposed (-.15) using Adobe Camera Raw 4.2
http://www.paperbag.no/images/ISO1600_20D.jpg

This is shot at ISO100 and overexposed (+2.40) using Adobe Camera Raw 4.2
http://www.paperbag.no/images/ISO100_20D.jpg

Clearly the images shot at high ISO is much cleaner than the image with added exposure in post. The images are from a Canon 20D.

It is my understanding now that the RED does the same as in picture 2. Can somebody with the knowledge and understanding please tell me why it doesn't do "gain up on the sensor" as in a DSLR? (At least that's what I think a DSLR does :umm: )

Please understand that this is a question I ask out of curiosity and a wish to understand the camera better. I'm not complaining! :innocent:

Regards
Fredrik Harreschou

[Kevin Halverson]

You are correct, the RED ONE has fixed analog gain. The image recorded is the same regardless of the "ISO" setting of the camera. The only impact that the "ISO" setting has it to effect the meta data, and the resulting default setting used in the post development tool, but this can be over ridden so the practical impact falls into the category of "insignificant".

If one didn't know the capability of the camera, one might jump to the conclusion that not having configurable analog gain is a problem. Rather, the 12 bit quantizers and nearly 12 bit dynamic range, makes this fact much less of an issue than it would on nearly any other camera.

[I Bloom]

You are correct, the RED ONE has fixed analog gain. The image recorded is the same regardless of the "ISO" setting of the camera. The only impact that the "ISO" setting has it to effect the meta data, and the resulting default setting used in the post development tool, but this can be over ridden so the practical impact falls into the category of "insignificant".

If one didn't know the capability of the camera, one might jump to the conclusion that not having configurable analog gain is a problem. Rather, the 12 bit quantizers and nearly 12 bit dynamic range, makes this fact much less of an issue than it would on nearly any other camera.

This is all correct but I'm thinking the 20D is also 12bit.

Frederick can you upload the uncorrected images from the camera as 16 bit TIFFs.

IBloom

[Kevin Halverson]

Just because something has a 12 bit quantizer path, it may or may not have an analog noise floor that can utilize anything near the full number of quantizer steps. The RED ONE is nearly perfect in this and exhibits a very usable 11.3 bits of dynamic range. I doubt that the 20D has the same level of performance despite a similar quantizer sample width.

[jbeale]

I'm in 100% agreement with khmuse. Another way to think about it is all those cheap PC soundcards that claimed "16-bit audio" and "CD quality" even though they really had all kinds of noise and spurious response that limited the true performance (record or playback) to something much lower than the 96 dB dynamic range that 16 bits gives you in theory.

I have a 20D too and it's a great camera, but I think your 1600 vs 100 ASA photos demonstrate it has some analog noise that is greater than what theoretically perfect 12-bit performance would be. Also, the new Canon EOS 40D apparently uses 14-bit A/D, but the testing I've seen indicates the noise level is about the same as the earlier 12-bit 30D and 20D.

[Graeme Nattress]

I did similar tests with a 20d and found analogue gain much superior to that of digital gain. With RED, I don't come to the same conclusion.

Graeme

[Fredrik Harreschou]

Thank you for your interest and replies.

IBloom, you can download the uncorrected tiffs (in zip files) here:
http://www.paperbag.no/images/ISO100_20D_16bit_tiff.zip and here http://www.paperbag.no/images/ISO1600_20D_16bit_tiff.zip

(I should mention that the images in the example in my last post were corrected with the Auto function in Camera Raw. That includes Shadows, Brightness, Contrast etc as well as Exposure.)

I guess I already knew that gain was fixed in the camera. I was just a little surprised at the difference. Honestly I shoot with my DSLR at ISO100 or 200 most of the time, so I haven't really examined high ISO vs Camera Raw exposure compensation that much.


If one didn't know the capability of the camera, one might jump to the conclusion that not having configurable analog gain is a problem. Rather, the 12 bit quantizers and nearly 12 bit dynamic range, makes this fact much less of an issue than it would on nearly any other camera.

I totally agree. The images speak for themselves. But does this mean that the RED sensor is pushed to its highest "acceptable level" (as in ISO1600/3200 on a DSLR)? I was thinking that if it had the same ability to "gain on the sensor" (sorry i have post production background) - would that be with less noise than exposing in post?

Is it because of heat issues that the gain is fixed in the RED? Or data rates? Or is it the compression algorithm? I guess a DSLR has a lot more time to do "pixel magic" with an image than the RED can do with a stream of the same resolution, at least that makes sense to me.

Again I'm just asking because I'm curious and that the answer maybe lead me to a new and better understanding on how to work with the camera - in six months time...:sad:

Another question:
I believe the sensor is rated at 320. How does this translate to ISO on a DSLR? I've been thinking that I could use my DSLR as a light meter, and with the right ISO setting, I hope I can use the 20D as a combined directors viewfinder and lightmeter, especially when scouting. Considering I will be using Canon glass and the crop factor is similar, I was thinking I could setup my DSLR to capture an image pretty close to what the RED would do. Any ideas on how to calculate this, or will it have to be tested to be any accurate?

Regards,
Fredrik

[Fredrik Harreschou]

I did similar tests with a 20d and found analogue gain much superior to that of digital gain. With RED, I don't come to the same conclusion.

Graeme

Graeme, thank you for your answer. You answered most of my questions before I posted them! I guess I spent an hour on that last post...

Very interesting. I was assuming this was something that applied to CMOS sensors in general.
PS. Since you have been testing with the 20D, maybe you know the answer to my question regarding the ASA rating of the 20D vs the RED?

Regards,
Fredrik

[Graeme Nattress]

Sorry, I've not done that ISO experiment.

Graeme

[Fredrik Harreschou]

Thank you again for your reply.

But is it something that could be established and be consistent so that it could be used reliably? It would be great to walk around at possible locations with my 20D and get quick feedback on available light.

And if I find a location I like, I will pick up my RED and shoot a second of 4K to grab some stills from. :clown2:

Regards,
Fredrik

[Graeme Nattress]

Possibly - but depends on the metering in camera, and the 20D doesn't have a spot meter, so you'd have to be careful.

Graeme

[Kevin Halverson]

I did similar tests with a 20d and found analogue gain much superior to that of digital gain. With RED, I don't come to the same conclusion.

Graeme

I am not positive (though I would assume) that most DSLRs apply analog gain when changing the base ISO at least over a portion of their range, but not necessarily the entire range. Considering that some camera's have a 5 stop range, this would mean an analog gain range of just over 30 dB. While certainly possible, I wouldn't be surprised to learn that at the higher settings, the ISO shift results from a combination of analog gain and post processing.

At least in the case of the RED, you do have absolute knowledge of what was occurring, where in many DSLRs, its a bit more of a black box.

[Deanan]

We've done extensive testing with every possible gain setting in combination with voltage biases, etc. etc. and continue to do so on the quest for pulling out better images from the sensor. As Graeme says, we came to the conclusion that the benefits from analog gain in this case was a wash. When you consider the disadvantages to analog gain, in the end it doesn't make sense.

Of course, that is subject to change :)

d

[Gunleik Groven]

Well, if you change it: Do not take away to make it work like it does!

:)