IQ3 100MP technical camera tests: color cast, mazing artifact, tiling issue, DR etc

[kdphotography]

While I appreciate these test results, is it not fair to say that the results will be rendered mostly invalid after the new patch for C1 is released tomorrow (if delivered as promised)? I.e. the 100 will improve somewhat?

I was wondering when someone would take note.

Lot of work here, but I certainly won't be using an old version of C1 Pro to process my IQ3 100MP files.



Patience.....patience....patience.

 

[voidshatter]

> Unfortunately the IQ250 is still not fully supported

Sorry, but how do you mean it is not fully supported?

If I look into the histogram of the deep shadow, I still see lots of gaps for the IQ250. If I convert by Capture One then these gaps get filled somehow. See #8 and #59 of this thread. I don't know what leads to the discrepancy here.

 

[voidshatter]

to process my IQ3 100MP files.

Congratulations on your new purchase! I thought you'd keep using your IQ180 :grin:

 

[Iliah Borg]

If I look into the histogram of the deep shadow, I still see lots of gaps for the IQ250. If I convert by Capture One then these gaps get filled somehow. See #8 and #59 of this thread. I don't know what leads to the discrepancy here.

IQ250 is a 14-bit camera, so the gaps are inevitable. To see those clear, you can switch off "RAW data postprocessing" for PhaseOne in RawDigger Preferences, Vendor Specific tab. For more detailed analysis we need a shot of ColorChecker in both .iiq and .dng formats, it would be great if you can provide those.

 

[voidshatter]

IQ250 is a 14-bit camera, so the gaps are inevitable. To see those clear, you can switch off "RAW data postprocessing" for PhaseOne in RawDigger Preferences, Vendor Specific tab. For more detailed analysis we need a shot of ColorChecker in both .iiq and .dng formats, it would be great if you can provide those.

I'm uploading the IQ250 DNG files in addition to the IIQ files into this folder here: https://www.dropbox.com/sh/fgircd45mbjf62r/AAC2evO6UQMAOxmQpLccyzmfa?dl=0

Please check back in about 2 hours time

 

[Iliah Borg]

I'm uploading the IQ250 DNG files in addition to the IIQ files into this folder here: https://www.dropbox.com/sh/fgircd45mbjf62r/AAC2evO6UQMAOxmQpLccyzmfa?dl=0

Please check back in about 2 hours time

From the part that is already uploaded, say 01719 .iiq/.dng pair: was Lens Correction Tool applied to the shot (Lens Correction, Purple Fringing, and/or LCC)?

 

[voidshatter]

From the part that is already uploaded, say 01719 .iiq/.dng pair: was Lens Correction Tool applied to the shot (Lens Correction, Purple Fringing, and/or LCC)?

From what I can see there is no LCC applied. No Chromatic aberration removal or lens correction has been applied either.

I still remember that Adobe Camera Raw and Capture One handles Phase One files differently. Apparently Capture One has all those manufacturer's tweaks under the hood for best results.

 

[torger]

There are calibration data tags also for the CMOS backs. They're obviously less important to decode and apply than for CCD backs, but they're still there. I don't think all sensor calibration tags for the CMOS backs have been reverse engineered, at least it's not done in DCRaw. This is probably causing the difference for the IQ250 files.

It's messy to handle the calibration data for us third parties as the MF companies introduce new tags with new formats with new sensors (as new sensors have new needs). I've reverse engineered for some Dalsa sensors for IIQ, and I've done it for some Kodak sensors in the Hasselblad 3FR format. But I haven't done it for CMOS neither 3FR or IIQ, but I know there are tags, just make a exiftool -v listing and you see them.

As there's virtually noone using open source software or independent third parties when it comes to $40k backs there's a lot of work for little reward doing this. The reason I did it for the Kodaks was that I needed it myself for my own H4D-50 back, and for the Dalsa I thought about getting such a back and needed proper support for it, but in the end I didn't get one. Another reason that it's not that rewarding is that the files usually look perfectly okay without calibration data applied, the calibration data just adds a tiny bit extra quality on top. For the Dalsas it was very important though as tiling usually was quite visible if the calibration data was not applied. For the CMOS there's no obvious reason to apply the calibration data, at least for us that shoot an LCC for each shot, but I don't know for sure as I don't know what's in all those tags.

In any case it's most likely that the CMOS tags are quite simple though as the calibration need is much smaller for CMOS than a CCD with external ADC.

 

[voidshatter]

There are calibration data tags also for the CMOS backs. They're obviously less important to decode and apply than for CCD backs, but they're still there. I don't think all sensor calibration tags for the CMOS backs have been reverse engineered, at least it's not done in DCRaw. This is probably causing the difference for the IQ250 files.

It's messy to handle the calibration data for us third parties as the MF companies introduce new tags with new formats with new sensors (as new sensors have new needs). I've reverse engineered for some Dalsa sensors for IIQ, and I've done it for some Kodak sensors in the Hasselblad 3FR format. But I haven't done it for CMOS neither 3FR or IIQ, but I know there are tags, just make a exiftool -v listing and you see them.

As there's virtually noone using open source software or independent third parties when it comes to $40k backs there's a lot of work for little reward doing this. The reason I did it for the Kodaks was that I needed it myself for my own H4D-50 back, and for the Dalsa I thought about getting such a back and needed proper support for it, but in the end I didn't get one. Another reason that it's not that rewarding is that the files usually look perfectly okay without calibration data applied, the calibration data just adds a tiny bit extra quality on top. For the Dalsas it was very important though as tiling usually was quite visible if the calibration data was not applied. For the CMOS there's no obvious reason to apply the calibration data, at least for us that shoot an LCC for each shot, but I don't know for sure as I don't know what's in all those tags.

It's most likely that the CMOS tags are quite simple though as the calibration need is much smaller for CMOS than a CCD with external ADC.

Yes see my post above - how Capture One calibrates the IQ250 when compared against ACR (presumably) without calibration.

Sometimes if you decode the IIQ of the IQ250 or Credo 50 by Adobe Camera Raw / Lightroom directly, shadow SNR gets beaten by the Nikon D810. In this regard, the Canikon solutions are less dependent on the RAW decoding software.

 

[Iliah Borg]

> From what I can see there is no LCC applied. No Chromatic aberration removal or lens correction has been applied either.

Thank you. RawDigger displays raw data and statistics; DNG is pre-processed (for example, the black level is subtracted, and in non-uniform manner), and that is normal for raw conversion. But RawDigger is not a raw converter, it is more or less a tool to discover the true nature of the raw data, and the tweaks and "cheats" added by different raw converters. If one wants to see the raw starting point to evaluate the sensor and lens performance, that's what RawDigger is for.

 

[torger]

Yes see my post above - how Capture One calibrates the IQ250 when compared against ACR (presumably) without calibration.

Sometimes if you decode the IIQ of the IQ250 or Credo 50 by Adobe Camera Raw / Lightroom directly, shadow SNR gets beaten by the Nikon D810. In this regard, the Canikon solutions are less dependent on the RAW decoding software.

Adobe probably gets tag documentation from Phase One, and I know they get it from Hasselblad (I've asked).

I know Adobe's DNG converter applies calibration data to Dalsa IIQs (so the DNG you get is "cooked"), but even then it was not exactly as Capture One. Some calibration data tags are things like polynomial curve handles so it can vary a few sample values depending on implementation. And at the time Capture One actually had a bug in their flat field calibration data application, so Adobe did it better at that particular point in time

With the IQ250 perhaps Adobe thought that calibration data made so little difference that it was not worth the effort implementing it even if they had documentation.

It should also be said that Capture One can in some cases make post-processing of the file, additional cooking, to improve quality. Removal of tile lines and micro lens ripple is one such post-processing thing that is not related to any calibration data. It's not unlikely they make additional minor tweaks, and indeed this is one thing Doug loves to speak about -- the advantage to be both manufacturer of camera hardware and raw converter in one. Using third party software you can miss out on some of the minor tweaks.

 

[voidshatter]

> From what I can see there is no LCC applied. No Chromatic aberration removal or lens correction has been applied either.

Thank you. RawDigger displays raw data and statistics; DNG is pre-processed (for example, the black level is subtracted, and in non-uniform manner), and that is normal for raw conversion. But RawDigger is not a raw converter, it is more or less a tool to discover the true nature of the raw data, and the tweaks and "cheats" added by different raw converters. If one wants to see the raw starting point to evaluate the sensor and lens performance, that's what RawDigger is for.

Or to be more precisely, to evaluate the starting point of the sensor and lens performance out of the camera, after the tweaks and "cheats" added by the camera's chip and firmware before it writes the RAW file into the memory card.

Which one could be better for tweaking and "cheating"? In-camera chip + firmware, or a sophisticated software like Capture One? I would imagine the latter in most cases.

 

[torger]

Calibration data usually fixes linearity, flat field (uniformity variations), tiling, mapping out dead (or half-dead) pixels. So if you evaluate those things your decoder must apply it of course. On the Hassy Kodak I looked at there was a strange type of calibration that moved some pixel data from the neighboring pixel, but it really was required to make the demosaicer work optimally (I guess it was some sort of readout crosstalk compensation). There's also tags like sensor temperature and such and I imagine that Capture One could be steering some pre-step noise reduction parameters based on those values.

So while evaluating raw data is valuable, one also needs to check the end result in the native raw converter too as software conversion is very much a part of the final image quality. Anyone who's used Capture One over a longer period of time knows that it happens now and then that they adjust tweaks, sometimes introducing bugs (like the LCC bug now).

Regarding tweaks I think the noise reduction part (the one that is always applied and can't be turned off) is probably what you gain most in using the native software, and indeed fine-tuned noise reduction has kept the "usable DR" of the medium format backs at a higher level than the sensors should actually be capable of. With third party software you can do good noise reduction too, but you generally need to tweak it manually and it's not always possible to make as fine tweaks as the native raw converts do. Noise reduction is not part of the calibration data, but some of the tags can be used as parameters into the noise reduction algorithm.

 

[ErikKaffehr]

Hi Void,

To my eyes the IQ3 100MP is quiet a bite more noisy than the IQ250. My eyes wrong? I have seen it Doug's round tower shot too.

Best regards
Erik

Now, the show time of the Sony CMOS sensor - dynamic range under long exposure!

First of all, it was lucky that I bracketed a bit for the long exposure shots so I have had just the right parameters for analysis!

Indeed, when Phase One wraps up the ISO settings (along with the dedicated long exposure mode for the IQ380 and IQ260) things can get very confusing - what is the real native ISO?

Instead of starting another long debate here, I employ a simpler method to try to minimize dispute:

I shoot each back at the lowest possible ISO (where for the IQ380 it is ISO 200 in the dedicated long exposure mode). I use the Lee Big Stopper. I shoot at the same aperture.

What I have found is that regardless of the difference of ISO among these three backs, indeed the same exposure time gave me about the same highlight headroom! See proof below: I converted all RAW files by Capture One v9.0.2 then checked the mean value of the fixed sky region. These are about the same.

Then I used the following settings for each image: I match the exposure of each image, then use the "Pick white balance tool" on the same spot, then push +100 shadow and pull +100 highlight, then disable noise reduction.

Note that for these two CMOS backs when I took the long exposure shot I disabled darkframe noise reduction by setting "Camera Mode" to "Aerial". This is indeed a key feature for landscape photography so to avoid waiting for the noise reduction countdown! Then for these three backs I moved the slider of "Single Pixel" to 1 (out of 100).

Now we look into the shadow: boom! You know what I'm talking about and why I am a fan of the Sony CMOS sensor! It is really the king in this kind of shootings! This is what I am willing to sacrifice other things for.

View attachment 115578

View attachment 115579

 

[Dan Santoso]

Round tower shots iso 1600 n 3200 is nosier than pentax iso 16000 if i recall...

If this new back only usable in iso 800 people will not upgrade from IQ180...

 

[torger]

Round tower shots iso 1600 n 3200 is nosier than pentax iso 16000 if i recall...

If this new back only usable in iso 800 people will not upgrade from IQ180...

Has high ISO become important for MFD users that quickly? You could always have an IQ250 laying around for those low light action shots

Maybe those 15 bits of base ISO DR couldn't be combined with great high ISO performance? It would be interesting if they've made a deliberate trade-off.

 

[ErikKaffehr]

Hi,

High DR is essentially also good high ISO capability.

Increasing ISO is simply underexposure, with noisy readout it can be helpful to increase analogue gain and that can be helpful for high ISO.

Best regards
Erik

Has high ISO become important for MFD users that quickly? You could always have an IQ250 laying around for those low light action shots

Maybe those 15 bits of base ISO DR couldn't be combined with great high ISO performance? It would be interesting if they've made a deliberate trade-off.

 

[torger]

I just made a quick visual inspection iso3200 compared to A7r2 on a per pixel basis. I couldn't see that IQ3100 was any worse, if anything it seemed a little cleaner. This in RawTherapee with no noise reduction tricks activated.

 

[Lobalobo]

If you don't shift that much then it's hardly visible on the CMOS sensors. For my real world usage I have only encountered this once and that was when I was too greedy to shift it to the extreme.

On the other hand, it is very easy to see that on the CCD sensors even unshifted, and that could be 3 vertical lines in addition. The 8 partitions are clearly visible on a fullframe CCD sensor by eye, but when you stare at the IQ3 100MP surface it looks like a whole (without any partitions at all).

IQ260+23HR unshifted:

View attachment 115644

IQ260+40HR unshifted but at ISO 200:

View attachment 115645

Have enjoyed this discussion and the technical education it provides (academic to me as I don't own a digital back, at least not yet). In this regard, I do have questions about the (lovely) long-exposure image of London. Are the tiling lines, which are clearly visible, shown to illustrate what the file looks like before they are removed in post, or are they not removable? If the latter, I'd be surprised as long-exposure is a feature of the IQ260, as I understand it, and it would seem to be asking a lot for a photographer to spend $40,000 on a piece of equipment so handicapped. Do these lines appear only in long exposure, or on shifting too? And, yes, I do understand that such lines may not be visible unless there is a uniform background or foreground such as sky or water, but as the London image shows, it is often desirable to have such backgrounds or foregrounds.

 

[Dan Santoso]

How is it compare to Pentax ISO 3200? Same Sony Sensor.....