• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

20D - ISO sensitivity values in Full Auto

D

Doug Kerr

Guest
I was surprised this morning to see a message from a member of dpr calling attention to the fact that when the EOS 20D is in Full Auto ("green box") mode, it can apparently adopt ISO sensitvities not in the list of ones we can select from (e.g., ISO 320).

I thought perhaps he was confused, but evidently not.

A quick test shooting out the front door in Full Auto mode brought into play sensitivities of ISO 320 and ISO 250.

I was way startled, to say the least.

What do we know about this?
 

Ray West

New member
Hi Doug,

There is no reason for any of the settings, iso, speed or aperture to comply with man made film values. Basically, I think we should throw all the current values away, and start again. (that could be taken in other ways, but similar to qwerty keyboard layout - we are stuck with it for the wrong reasons).

In film, pita to change iso on the fly, in digital, no prob. How do you want to control it? sliders, set any value? nice if you have a decent preview facility.

At the pro end, all the film baggage is being dragged into the digital camera design - about time folk started thinking 'out of the box'. Maybe the mobile phone camera end has the right ideas (note 'maybe')

Best wishes,

Ray
 

Ray West

New member
Hi Harvey,

I know, I was just off on one. If the camera is in auto, it can set whatever values it likes within range of its capabilites, since it is a leaf or similar type aperture, a variable speed shutter, and whatever amplifier settings for iso. Once it has sorted out what it wants to set, I suspect it displays what it has decided on. There is no need for it to decide on settings that are available in manual mode, for a number of reasons. In fact it could be that if it sets iso 145 it displays iso 150. I was trying to make a particular point wrt how we (folk in general) perceive things, based on what has gone before. There is no need for a digital camera in auto mode to restrain itself to whatever may be displayed for setting in manual mode.

Now, if we decide we want infinitely adjustable manual settings, is that a good thing, or bad? What is the control interface requirements, how is it displayed? How does this effect the price of fish??

I do not have much time to go into detail on my thoughts on this (even if they mattered, and if they did matter, I would not be voicing them on an open forum.)

It was wrongly attributed to JFK, but a useful quote is something like 'You see things that are, and ask why, I see things that are not, and ask why not' (something like...) (Just in general, not aimed at anyone here)

Best wishes,

Ray
 

Nill Toulme

New member
Very interesting rambles Ray. ;-) Nevertheless, one of my few peeves about the 20D (the other two being the firecracker-loud shutter and the gimcrack CF door) is the inability to set ISO between full stops. The fact that it can do so on its own motion suggests to me that a firmware hack is in order. Anybody volunteer?

Nill
~~
www.toulme.net
 
D

Doug Kerr

Guest
All well and good but

All that is well and good, but I thought it had been pretty well established that the analog pre-digitization amplifiers in these specific sensors can only have their gain set to a finite repertoire of gains, not including those that would be needed for "intermediate" ("1/3 stop") ISO sensitivities.

There is considerable evidence that the "1/3 stop" ISO sensitivites in the EOS 5D are achieved not by changing pre-digitization amplifier gain but by mathematical manipulation of the digitized output (a process that has some bad side effects).

So it would be a bit surprising to learn that this game has been practiced all along in the 20D, but only in Fulll Auto.

Best regards,

Doug
 

Asher Kelman

OPF Owner/Editor-in-Chief
Guys,

Have you seen any evidence of higher ISO values being rated lower than it is in reality in the 20D? ie 1600 being 3200 to 6400!

Asher
 
D

Doug Kerr

Guest
Hi, Asher,

If we start with the basic "sanity check" for digital EOS exposure metering systems recommended by Chuck Westfall and work back through the labyrinthine trail of the various pertinent ISO standards, one concludes that the ISO senstitvity ratings on EOS digital cameras are about 75% the values that would be rated under the applicable ISO standard. That is, when we set a sensitivity of ISO 100, the behavior of the sensor system would probably lead to a rating, under the ISO defnition and associated test procedure, of about ISO 135.

I conjecture that this comes from the following:

1. Canon wishes to use an expsure metering strategy that has less "headroom" than in the one that would result from following the ISO standard for expsure meters. (That is, it "exposes more to the right".) This is probably justified on the basis that the "evaluative" metering system is smarter about the range of scene luminance than the "dumb" metering system on which the ISO scheme is predicated, and thus needs less "headroom" to protect against highlight blowout.

2. If they did this by just shifting the calibration constant of the metering system, then phtographers using (properly calibrated) external expsure meters would get indications inconsistent with the indications of the camera's metering system, a sure cause of consternation.

3. The same end result can be attained (surreptitiously) by "understating" the ISO sensitivity of the sensor system. This causes the metering system to call for a greater expsure that would be called for under the orthodox ISO scheme, even with "orthodox" exposure meter calibration.

Of course, far be it from me to divine The Mind of Canon - I'm just a telephone engineer.

Best regards,

Doug
 
Last edited:

Roger Lambert

New member
If the ISO values as stated in the Full Auto exif are indeed true, then does it not argue for - gasp- the concept that Full Auto may in fact yield superior results to other more manual settings during certain photographic conditions?

As much as we hate to admit it, these are very sophisticated cameras, with very powerful processors, capable of making nearly instantaneous evaluations of meterings and the calculations needed to optimize optimal exposure.

Perhaps the art may lie in our judgement of when to use the Full Auto function, or perhaps when to utilize as much of the camera's automated capabilities as are genuinely needed. And not to let our ego interfere with photograhic excellence.

How many of use have actually tested or used the Full Auto function to it true capability?

I know I am guilty of this, despite the fact that on occasion I have been flabbergasted to discover that the Full Auto clearly outperformed my best efforts under difficult conditions. I chalked it up to my neophyte status and determined to improve my skills.

But perhaps this is a fools errand. And it is a wiser course to use all the tools for which we have paid Canon so handsomely?
 
D

Doug Kerr

Guest
Hi, Roger,

If the ISO values as stated in the Full Auto exif are indeed true, then does it not argue for - gasp- the concept that Full Auto may in fact yield superior results to other more manual settings during certain photographic conditions?

Indeed! Scary! Could almost make one give up one's ideological beliefs!

Best regards,

Doug
 

Asher Kelman

OPF Owner/Editor-in-Chief
Doug Kerr said:
Hi, Roger,



Indeed! Scary! Could almost make one give up one's ideological beliefs!

Best regards,

Doug

Yes,

but we'd better get used to it! As the technology of the sensor and data handling improves, we will have essentially "B1-bomber" class cameras, fly them manually at your own risk.

Concepts like ISO will be outdated once there will be local pixel by pixel control of filling of pixel wells or combining adjacent pixels to get an accurate number.

This methodology is just around the corner with Canon, HP and others about to bring us this (probably first) in digicams. Already Texas instruments has provided Kodak and HP with a smart chip to enable mass produced digicams to take pretty great pictures in scenes with shadow and sunlight.

There will always be a place for manual artistry.

The creativity of photographers, however, is ultimately in the subject, setup, choices for inclusion, lighting and producing a print that is outstanding.

Each camera we now buy is obsolete at purchase, but very wonderful.

The key is to buy well enough to make money and art that makes money or just pleasure!

Asher
 

Alan T. Price

New member
Aren't these just 1/3 stop increments in line with the custom function 6 ?
Admittedly we can't normally set a 1/3 stop increment ISO but that doen't mean the camera can't.
 

John Sheehy

New member
Roger Lambert said:
If the ISO values as stated in the Full Auto exif are indeed true, then does it not argue for - gasp- the concept that Full Auto may in fact yield superior results to other more manual settings during certain photographic conditions?

It's a gamble, actually. For the 20D to generate 1/3-stop ISOs, it has to push and pull by 1/3 stop, so the JPEGs from the greenbox mode could have more or less shadow noise than the main ISOs. The 160/320/... group will have less shadow noise, and the 125/250/... group will have more.
 
Nill Toulme said:
Yikes... is that true of the 1-series also Doug?

From my testing I concluded that my 1Ds Mark II only uses ISO 100, 200, 400 and 800 as pre-quantization gain settings. The other intermediate and higher ISOs are postprocessed gain on underexposed image data, with somewhat more noise.

ISO 50 is a special case, with its data being overexposed (losing a stop of headroom) and pulled a stop in postprocessing, with better noise.

I don't know about the 1D Mark II, but I presume something similar.

Bart
 

John Sheehy

New member
Bart_van_der_Wolf said:
From my testing I concluded that my 1Ds Mark II

Have you ever looked at the standard deviation of noise in an ISO 100 blackframe from that camera? I have been very curious to figure out if some of the figures I've seen make sense. Roger's web pages imply 1.16 and 1.26, depending on how you read them, but that is so much lower than any of the other Canons, even the 5D, that I am skeptical. The astrosurf website implies something similar for the 5D, yet the 5D clearly has a standard deviation of about 1.95 in an ISO 100 RAW blackframe. The discrepancy is one that would be accounted for if the measurement was taken from data that was clipped at the blackpoint, but standard deviations taken from black-clipped blackframes are meaningless.

The easy way to do this, if you have IRIS, is to just load the RAW file into IRIS, and type the "stat" command and look at the sigma value. You can multiply the whole image by 10 or 100 first for more precision, as the sigma value given does not have much precision.
 

John Sheehy

New member
John Sheehy said:
Have you ever looked at the standard deviation of noise in an ISO 100 blackframe from that camera?

Sorry. I missed the little 's' the first time. I'd still be interested in the blackframe noise from the 1DsmkII, though.

Does anyone have the 1DmkII and IRIS?
 
John Sheehy said:
Have you ever looked at the standard deviation of noise in an ISO 100 blackframe from that camera?

Yes, I am currently assembling the data for a more formal presentation of the results (a tedious job of reading multiple files and collecting derived statistics in a spreadsheet for analysis). And there is more I'd like to add later (e.g. Imatest single image comparison to the multiple image method used by Roger an others, both methods have their benefits/drawbacks IMHO). I also need to check if the Astrosurf's IRIS data relate well to the ImagesPlus data (basically a test between their capabilities to extract totally unprocessed Raw data, needed for such fundamental data crunching), and if the L and H settings are anything else than postprocessed.

I have been very curious to figure out if some of the figures I've seen make sense. Roger's web pages imply 1.16 and 1.26, depending on how you read them, but that is so much lower than any of the other Canons, even the 5D, that I am skeptical.

I just initiated a private email exchange with Roger to clarify some anomalies I found. We both use 'ImagesPlus' as evaluation software (which uses the official Canon SDK) so comparing results should be easy.

The astrosurf website implies something similar for the 5D, yet the 5D clearly has a standard deviation of about 1.95 in an ISO 100 RAW blackframe. The discrepancy is one that would be accounted for if the measurement was taken from data that was clipped at the blackpoint, but standard deviations taken from black-clipped blackframes are meaningless.

ImagesPlus allows to extract the unprocessed/non-White-balanced/non-gamma-adjusted data before demosaicing. That allows to very accurately analyze the sensor data. Sofar, there is no clipping or other processing other than what the sensor recorded that I can detect. That data even allows to derive the per channel ADC gain used, which can be useful in determining the native camera ISO.

The easy way to do this, if you have IRIS, is to just load the RAW file into IRIS, and type the "stat" command and look at the sigma value. You can multiply the whole image by 10 or 100 first for more precision, as the sigma value given does not have much precision.

One IRIS issue I encountered with my current limited memory box, is that 1Ds Mark II files exceed the IRIS memory allocation for my modest hardware (I intend to up-grade/replace the box next year by something more potent). I'll have to run a cross-check with my 20D to see if those files produce similar results between IRIS and ImagesPlus.

For the moment, this is part of my EOS-1Ds Mark II noise analysis based on black-frames (1/8000th of a second exposures, to minimize dark current noise, with lens and viewfinder covered), scaled to 16-bit Data Numbers (from 12-bit Sensor numbers):
ReadNoise_1Ds2.png


It reveals that also the 1Ds2 uses full stop analog amplification with ISO settings, and that ISO 800 might be the highest ISO that uses analog gain. Higher/in-between ISO settings seem to be post-processing of already quantized data. That would suggest better noise results for several higher ISO settings than lower 'in-between' settings. Obviously my results are based on a sample population of 1 unit ...

Bart
 

John Sheehy

New member
Bart_van_der_Wolf said:
Yes, I am currently assembling the data for a more formal presentation of the results (a tedious job of reading multiple files and collecting derived statistics in a spreadsheet for analysis). And there is more I'd like to add later (e.g. Imatest single image comparison to the multiple image method used by Roger an others, both methods have their benefits/drawbacks IMHO). I also need to check if the Astrosurf's IRIS data relate well to the ImagesPlus data (basically a test between their capabilities to extract totally unprocessed Raw data, needed for such fundamental data crunching), and if the L and H settings are anything else than postprocessed.

From one of the last correspondences I had with Roger on usenet, it seemed that ImagesPlus was doing something wrong, at least for some Nikon RAWs. Iris is the only program that I feel comfortable with; all the others seem to do things behind the scenes to alter the RAW data. DCRAW now has a "-D" switch that outputs actual RAW, allegedly, but unfortunately, the only program I have that reads 16-bit files literally is IRIS, and I can load directly into IRIS so DCRAW is out of the picture. If you output a PSD in DCRAW, it is pretty useless in PS as it is posterized by PS' pseudo-16-bit mode.

I just initiated a private email exchange with Roger to clarify some anomalies I found. We both use 'ImagesPlus' as evaluation software (which uses the official Canon SDK) so comparing results should be easy.

I wouldn't trust the Canon SDK. I don't think it delivers true RAW data. Canon seems to deny the existence of the top 1/3 stop of RAW highlights, and clips away the highlights that surpass that level in the red and blue channels after color balancing, as well. It may black-clip as well.

For the moment, this is part of my EOS-1Ds Mark II noise analysis based on black-frames (1/8000th of a second exposures, to minimize dark current noise, with lens and viewfinder covered), scaled to 16-bit Data Numbers (from 12-bit Sensor numbers):

It reveals that also the 1Ds2 uses full stop analog amplification with ISO settings, and that ISO 800 might be the highest ISO that uses analog gain. Higher/in-between ISO settings seem to be post-processing of already quantized data. That would suggest better noise results for several higher ISO settings than lower 'in-between' settings. Obviously my results are based on a sample population of 1 unit ...

The reverse-triplet pattern is pretty safely assumed to be design, I would think. A variation of the values from camera to camera would reflect differences of individual specimens.

Why is the noise different for different channels? Raw noise is colorblind, until WB is performed. If you have rendered a 16-bit linear TIFF with the Canon SDK, I think the color space is still interpolated, so that would not be a good way to measure native response. Saturations are changed based on hue, and hues are shifted, based on hue in RAW conversions.

I am less skeptical about the low noise now, as I picked up a 400D as a second camera for wide lenses (I shoot mostly telephoto and don't like changing lenses in the field, and I may need to send my 20D in for servicing, and the 10D is not a suitable substitute for low light or EF-S lenses). The 400D has less readout noise than the 5D at ISOs 100 and 200, with more dynamic range (at the pixel level). The standard deviations are 1.65 and 1.84 (100 and 200, respectively), so the figures of about 1.3 for the 1-series cameras for 100 don't sound as far fetched now. I have to wonder why the 5D has so much readout noise (1.95 and 2.05, respectively), relative to the 1 series. Perhaps, until now, readout noises much lower than 2.0 were expensive. Hopefully, the 30D successor will have lower readout noise for low ISOs.
 
Top