MF CMOS -- now it gets interesting

[synn]

IQ250 sample images...

https://www.google.com/search?q=IQ250+site%3Aphaseone.com

Better link -

http://gallery.phaseone.com/#/Action-sports/Tim-Kemple/Tim-Kemple_1_gallery.jpg

note that the camera is listed as the 645DF+, so perhaps nothing new on that front coming just yet.

Probably early days to talk about it, but the color rendition does look different to the older phase backs.

Regards,
Sandeep

[ErikKaffehr]

Hi,

Live view can be nice on tech cams and a couple of stops more shadow detail wouldn't hurt either.

Best regards
Erik

High ISO isn't important for tech cams so they can stick with CCD

[Bart_van_der_Wolf]

Probably early days to talk about it, but the color rendition does look different to the older phase backs.

Hi Sandeep,

Indeed, we'll have to wait and see what they actually do with their choices of Bayer CFA filters. Remember that many say the Leaf backs render skin color much better than other MF backs, yet both are based on CCDs. That shows that it is not the sensor (which just collects photons), but the other choices that make a difference.

Cheers,
Bart

[Theodoros]

2 24mp sensors side by side make 48x36, add a little marketing magic and you have 50mp. I can't see them releasing a 44x33 after it took years to use the full frame size.

It may be a new sensor after all with mind of alternative sizes to be made… Lets wait and see, in a few days we'll know, certainly sensor size along with price can tell a lot for market positioning, if price is low it should lead to a conclusion that the target is to "bridge" DSLRs with respect to CCD MF… if the price is near CCD backs, it should suggest that they consider the Cmos backs as competitive/alternative solutions to the current CCD backs.

[free1000]

IQ250 sample images...

https://www.google.com/search?q=IQ250+site%3Aphaseone.com

Better link -

http://gallery.phaseone.com/#/Action-sports/Tim-Kemple/Tim-Kemple_1_gallery.jpg

note that the camera is listed as the 645DF+, so perhaps nothing new on that front coming just yet.

Ha, picture appears gone.  Did someone get their botty spanked?

[hjulenissen]

Hi Sandeep,

Indeed, we'll have to wait and see what they actually do with their choices of Bayer CFA filters. Remember that many say the Leaf backs render skin color much better than other MF backs, yet both are based on CDDs. That shows that it is not the sensor (which just collects photons), but the other choices that make a difference.

Cheers,
Bart

If the "color difference" is only due to different choices in CFA, how much of that difference can be compensated through the color correction that is applied in both cases?

If you use something like the xrite colorchecker passport on a MF CCD camera and a FF CMOS camera, and process them similary in rawtherapee or Lightroom, how different are the colors?

-h

[synn]

If the "color difference" is only due to different choices in CFA, how much of that difference can be compensated through the color correction that is applied in both cases?

If you use something like the xrite colorchecker passport on a MF CCD camera and a FF CMOS camera, and process them similary in rawtherapee or Lightroom, how different are the colors?

-h

It's not just a difference in color rendering.
Pretty much every properly shot MF file that I have seen picks up subtle shade differences of the same base color in a way 35mm files don't. No amount of color chart shooting and calibration can fix that. Believe me, I've tried.

Is it a CCD vs CMOS thing? I dunno, I am too tired to fight that battle again.

All I know is, those sample images (That have now disappeared) had a more "Upscaled D800" look to them than being close to say, a P40+.

[Bart_van_der_Wolf]

If the "color difference" is only due to different choices in CFA, how much of that difference can be compensated through the color correction that is applied in both cases?

There is no exact number that describes it. The analog Bayer CFA bandpass of the subject spectral colors is quantized in 3 numbers. Those 3 numbers cannot be inverted back to the exact same full spectral information. Information was lost in the quantization and reduction of data. We can do a pretty convincing guesstimate of what the colors probably looked like, but some trade-offs remain to be made in that process, and inaccuracies will exist. 

If you use something like the xrite colorchecker passport on a MF CCD camera and a FF CMOS camera, and process them similary in rawtherapee or Lightroom, how different are the colors?

Everything else being equal? The same assumptions in reconstruction of the photon counts will result in the same output. There will be differences in e.g. noise though, so there will be small differences, with CMOS behaving better in e.g. the read-noise and higher ISOs, and lower thermal noise (dark current).

Cheers,
Bart

[Doug Peterson]

Ha, picture appears gone.  Did someone get their botty spanked?

You didn't see nothin .

[ErikKaffehr]

Hi,

An answer to your question may be this:

P45+ to the left and Sony Alpha 99 to the right.

But other posters may come up with other answers.

My full article is here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition

The article needs some rewrite for clarity.

If you use something like the xrite colorchecker passport on a MF CCD camera and a FF CMOS camera, and process them similary in rawtherapee or Lightroom, how different are the colors?

-h

[hjulenissen]

Hi,

An answer to your question may be this:

But other posters may come up with other answers.

My full article is here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/79-p45-colour-rendition

At a quick glance, the color charts does not immediately appear to favour one of the cameras?

-h

[hjulenissen]

Everything else being equal? The same assumptions in reconstruction of the photon counts will result in the same output. There will be differences in e.g. noise though, so there will be small differences, with CMOS behaving better in e.g. the read-noise and higher ISOs, and lower thermal noise (dark current).

With two different CFA responses, the output will generally not be equal. What I am requesting (and what Erik seems to have done) is a real-world test where one tries to remove the big unknown of MF raw developer color choices, and rather see what the developed raw files look like respectively when both attempt to color correct against some agreed upon reference.

Some (trivial) differences in color filter response may be easily fixed by a small change to the color correction matrix. Other differences might be impossible to solve without amplifying noise too much (inverse matrix going bananas). Some differences may be best seen for certain shades.

-h

[ErikKaffehr]

Hi,

The Sony was a bit more accurate, what that is worth.

Best regards
Erik

At a quick glance, the color charts does not immediately appear to favour one of the cameras?

-h

[Ken R]

Kids, let's not get too technical!

Erik, sometimes I wonder why you purchased the P45+ in the first place and why don't you sell it already. The A99 seems to be the perfect camera for you (The D800E and the A7R should also)

The Hasselblad and the IQ250 most likely won't have a "full frame" sensor from the looks of things. Smart money says the 50MP CMOS sensor is going to be 1.1x (36.8x49.1) or 1.3x crop (33x44 approx.), not an issue at all for most people though. Also being CMOS the sensor probably won't play well with wide angle tech camera lenses. Just like the A7R does not deal well at all with Leica M wide angle glass. (Rodenstock HR lenses should be ok, can't wait for tests!)

Also, one big question mark is the price point. A lot of folks here want the new Hasselblad (and IQ250) to be priced at Pentax 645D levels or close. Seriously doubt that is going to be the case although that would do WONDERS to increase the customer base of digital medium format.

Oh, and one more thing. The IQ250 will most likely be the better implementation of the 50mp CMOS sensor given PhaseOnes history of making the best of the sensors available and their trajectory of building very very solid backs with outstanding functionality (class leading for sure) 

[torger]

There's more to medium format than image quality. Even if I'd desire to have the DR of a Sony Exmor sensor the image quality of my ancient Aptus is certainly more than adequate, and then I get to use my tech cam which provides a shooting style I prefer. There's also quite fun and interesting to use legacy camera systems like Hassy V and Mamiya RZ, even if you not believe in the inherent superiority of the CCDs or whatever.

[Paul2660]

"High ISO is not important for a Tech camera"

High iso is can be very important for a tech camera.  If you are an outdoor shooter and working with wind.  Many times I have shoot combinations to stop the movement of my subjects, trees, flowers, as I don't prefer blur.  

Actually the Tech camera may make this more an issue if anything since many of the lenses available are best used in the F8 to F11 range.  In low light, (cloudy, or grey winter day) it can be very hard to get 1/60 to say 1/125th of a sec at base ISO of 50 or even 100.  You need to stay in the same aperture range of F8 to F11, you have a CF on the lens which takes yet another 2 to 2/5 stops of light.  So having a useable high iso of say 400 or 800 would be a real plus for me.

You can just as easily get into the need for a ISO on a good day, with sunlight but still windy.  Many times on stream treks I will have great light, but 15mph to 20mph gusts.  Here again I am going to shoot a bracket for the water (slow) but try to stop the tree branches.  ISO 50 or 100 again won't get you to the shutter speed you need even pulling off the ND or CLPL I might have.   You need to be at 1/125 at least to stop the motion clean and this again needs around ISO 400 minimum.   ISO 400 on a current high MP Phase back can be tricky, even with the IQ260 in my experience.  Plus you get back to the issue of "is it really iso 400 or 800" due to the fact it's a CCD instead of CMOS.  

Sensor plus does a great job here i in creating a noise free file, but the overall loss in resolution is not something I want.  

Paul Caldwell

[jerome_m]

Given the evidence that two companies making sensors for customers (like the military and satellite makers) for whom sensor cost is not much of a barrier, I doubt that increased cost is the reason why CCD's silt such at read noise command to CMOS sensors. Do you care to comment on the fundamental technical advantages of active pixel CMOS sensor design such as early amplification that active pixel CMOS sensor designs offer?

I don't understand what you want me to comment especially in the context of the present discussion.

All I am saying is that the new cmos backs will not have massively improved high iso compared to the present ccd backs. An educated guess is that the new technology gains about one stop sensitivity at the same sensor noise level. It is better than nothing, but it is not earth shattering.

We have people here complaining that iso 200 is the most they will use on their 50 mpix back (without micro-lenses) before the quality is degraded too much for their needs -> they will be able to use iso 400. Not more. We also have people who feel iso 800 is the max, they will use iso 1600, etc... The increase in noise level is going to be roughly one stop.

It is not much, especially when one considers that MF lenses are 2 to 3 stops slower than 24x36 primes.

[eronald]

If the "color difference" is only due to different choices in CFA, how much of that difference can be compensated through the color correction that is applied in both cases?

If you use something like the xrite colorchecker passport on a MF CCD camera and a FF CMOS camera, and process them similary in rawtherapee or Lightroom, how different are the colors?

-h

If the CFA is bad you lose color precision, and cannot recover it in post.

At random, I'd say two things happen:
- The CFA does not render the cone functions properly, so the numbers it spits out are wrong. (Luther-Ives issues)
- The CFA has been tweaked to let too much light through, and so loses color precision. (orthogonality issues)

Edmund

[hjulenissen]

If the CFA is bad, non- orthogonal you lose color precision, and cannot recover it in post.

I think that your reply is over-simplified. It should be easy to specify a filterbank that is orthogonal while still being colorimetric bonkers. I'd expect that any filterbank response that can mimic the desired response through a "well-behaved"*) 3x3 color correction matrix would be a good starting point.

The spectral dimension seems to be highly similar to the spatial dimension in this respect. You might say "if you choose a small aperture, you loose spatial details and cannot recover it in post". Except good sharpening/deconvolution can actually bring back details as long as they are still sufficiently large compared to the noise floor (and the PSF can be estimated).

So the question really is (perhaps)
* "how far from the (colorimetric) ideal spectral sensitivity is the color filter array"
* what does the colorimetric ideal color correction matrix look like
* at what ISO/signal levels can this matrix be applied before the output starts to appear "noisy"

It is my impression that Sony (and the MF guys) tends to prioritize accurate color, while Canon and Nikon prioritize luminance sensitivity.

-h
*)An identity matrix being good, but other forms that does not have large off-diagonal terms might also do.

[eronald]

I've amended my post a bit.
My knowledge of color is purely practical although I've written camera profiling software.
And my mathematics is usually pretty schematic until I take the pains to read the details on a subject.
Somebody who actually designs CFAs  would be the right person to answer your questions.

You can always end up with some real-world spectrum that will make a mockery out of your attempt to simulate the human eye. The colorimetric approach of collapsing a spectrum into 3 numbers means integrating a set of simulated observer functions, and if there are acute peaks in the spectrum things are going to go irrecoverably wrong very quickly.

I don't know whether trichromatic colorimetry based on a "standard observer" is such a good idea, after all people around the world have different eye colors and shoesizes, and one certainly shouldn't take it as trichromatic colorimetry as the ground truth of color perception. The original experiments on which colorimetry is based are very old.

Actually there are quite a few tetrachromats floating around in the human population, about 50% of the women and 8% of the men, although it is unclear what proportion are functional.

Edmund

I think that your reply is over-simplified. It should be easy to specify a filterbank that is orthogonal while still being colorimetric bonkers. I'd expect that any filterbank response that can mimic the desired response through a "well-behaved"*) 3x3 color correction matrix would be a good starting point.

The spectral dimension seems to be highly similar to the spatial dimension in this respect. You might say "if you choose a small aperture, you loose spatial details and cannot recover it in post". Except good sharpening/deconvolution can actually bring back details as long as they are still sufficiently large compared to the noise floor (and the PSF can be estimated).

So the question really is (perhaps)
* "how far from the (colorimetric) ideal spectral sensitivity is the color filter array"
* what does the colorimetric ideal color correction matrix look like
* at what ISO/signal levels can this matrix be applied before the output starts to appear "noisy"

It is my impression that Sony (and the MF guys) tends to prioritize accurate color, while Canon and Nikon prioritize luminance sensitivity.

-h
*)An identity matrix being good, but other forms that does not have large off-diagonal terms might also do.