IQ3100 Technical Camera tests, from Capture Integration

[Paul2660]

Lots of stuff on this link.  Appears that C1 9.0.3 will be needed to really work the files, due to the known issue of green cast on LCC conversion.  But a lot of downloadable files.  Indoor shoot only from what I can tell, but a lot of files to work with.

Here is the link:

https://captureintegration.com/phase-one-iq3-100mp-technical-camera-testing/

Paul C

[torger]

Thanks for the link. Lots of files to look at. This will take time to digest

[torger]

I've made an initial look with my tools, and it looks quite good.

The back seems to be more robust for horizontal shifts in landscape orientation (I think the IQ250 was the same) so it's not really a fully conclusive test. Looking at the shift amounts they have tested it seems like they have been very aware of the limitations, or found the "limits" with trial-and-error as the test shots do have more horizontal shift than vertical rise/fall. I would guess that past the provided max values in this test there's a sharp falloff in performance.

I can see with my tools that there is crosstalk in the files, but I really need some established reference to say something useful if it's much or not, it's too long ago I was deep into it. It's not so huge that you get any obvious demosaicing failures in any case, but the files are often quite underexposed (noisy) with soft details so it's much harder to evaluate than Doug's library test with the IQ250.

[ErikKaffehr]

Hi,

More than I ever would have asked for but I would not complain!

Thanks CI!

Best regards
Erik

Lots of stuff on this link.  Appears that C1 9.0.3 will be needed to really work the files, due to the known issue of green cast on LCC conversion.  But a lot of downloadable files.  Indoor shoot only from what I can tell, but a lot of files to work with.

Here is the link:

https://captureintegration.com/phase-one-iq3-100mp-technical-camera-testing/

Paul C

[Paul2660]

I would have liked to see 15mm of shift on the 32mm.

Paul C

[Chris Barrett]

Paul, I'm envisioning a crazy rig... Arca MF2 / Dex focal plane shutter / IQ3 100 / 'Blad CFi/e glass.  Crazy I tell you.  Id just need an ultra-wide solution.  I'm going to test my 'Blad glass with the new back on the XF Wednesday.  No shifts, but at least I can judge sharpness of the lenses.

CB

[jng]

Paul, I'm envisioning a crazy rig... Arca MF2 / Dex focal plane shutter / IQ3 100 / 'Blad CFi/e glass.  Crazy I tell you.  Id just need an ultra-wide solution.  I'm going to test my 'Blad glass with the new back on the XF Wednesday.  No shifts, but at least I can judge sharpness of the lenses.

CB

Chris,

I would be very interested in seeing your results from shooting the old V system glass on the 100 Mp back. In my experience, the 40/4 IF CFE renders beautifully on the IQ160 (very sharp) and some of the other old lenses perform quite nicely as well. I've pasted in an example below, shot with the 40/4 IF CFE @ f/11 on a 501CM so no shifts or tilts. You can go to the image on my Flickr site if you'd like to see the full size jpg (only cropped out the lower 1/3 of the image to make it square). I used default sharpening and mild lens-specific sharpening in C1 (yes there's a profile for this lens in C1!), and then mild sharpening in CC. The detail of the upper part of the tower holds up pretty well, IMHO.

- John

CF003498_C1 by John Ngai, on Flickr

[torger]

I've attached a few measurements.

It shows:
* IQ3200 40HR shifted right 15mm - from CI tests
* IQ3200 32HR shifted up 10mm and right 10mm - from CI tests
* IQ250 32HR shifted left 15mm and up 15mm - from Doug's library test
* IQ260 32HR shifted left 15mm and up 15mm - from Doug's library test

There are two heat maps per combination.

The first shows the signal loss, that is how many stops the channel (red green or blue) that vignettes most loses. If it's green it's less than one stop, yellow less than 4 stops, red more than 4 stops. That is if you have 15 stops of DR and loses 4 stops in the corner, you have 11 stops left there.

The second heat map shows crosstalk. Unfortunately crosstalk cannot be that accurately measured as the results depends on the light. The more similar red and blue content in the illuminant the lower crosstalk numbers, it does not mean that there is less crosstalk but only that the measurement can differ between what's crosstalk and what's real signal (and that is the problem that makes crosstalk so hard to correct for).

As the light probably has not been 100% the same between Doug's library test and CI's test it's not 100% comparable, but my guess is that it's about the same.

Anyway, the green means less than 1.5% which is normal variation, yellow less than 1.5-10% crosstalk, and red is 10%-30%

A few comments of the results.

* IQ3100 loses a bit more signal than IQ260, but it's probably compensated in full with the better DR of the sensor.
* IQ250/IQ3100 has the same crosstalk behavior (you see the same pattern, if they had been shifted same amount the heat maps would look very similar, minus the difference in sensor size), the IQ3100 may/might have a bit less crosstalk but not with much.
* The 40HR with 15mm right shift has negligible problems. I shall look at larger shifts later.
* The Sony sensors holds up better sideways and up/down concerning crosstalk.
* The IQ260 which is shifted to hard vignette in the example would have less than 5% worst case if not hard vignetting for the 32HR, ie not an issue.

From these test resluts it seems like the IQ3100 will yield similar results as to the IQ250 regarding shifting, possibly a little better but no big difference.

I think 10mm rise/fall is too little for this sensor size, I think one should push the tests to at least 12-15mm.

[torger]

Added CI's 40hr with 18mm left and 10mm up.

As you can see the crosstalk pattern is always the same, but here it only reaches 12% in the worst red zones while the 32hr pushed it to 24%.

How much is acceptable? That's up to you. Crosstalk means that red green and blue is mixed, that is tonality performance is reduced (colors gets gradually more desaturated and similar), and as green1 and green2 starts separating you eventually get demosaicing failures too (mazing).

For me personally I think I could live with the 40hr performance and shift without thinking, while the 32hr would feel more like a center frame lens, you will risk tonality when shifting it by any sizable amount. However as it looks okay in most cases I think many will do sizable shifts with it anyway.

This is obviously about shoehorning in a digital back that was not designed for these lenses, and we are pushed to compromise one of the classic key features of medium format -- superior tonality -- in the red zones it becomes probably worse than your mobile phone camera in that aspect, but the center is good and the CMOS feature set is of course hard to resist...

[voidshatter]

Added CI's 40hr with 18mm left and 10mm up.

As you can see the crosstalk pattern is always the same, but here it only reaches 12% in the worst red zones while the 32hr pushed it to 24%.

How much is acceptable? That's up to you. Crosstalk means that red green and blue is mixed, that is tonality performance is reduced (colors gets gradually more desaturated and similar), and as green1 and green2 starts separating you eventually get demosaicing failures too (mazing).

For me personally I think I could live with the 40hr performance and shift without thinking, while the 32hr would feel more like a center frame lens, you will risk tonality when shifting it by any sizable amount. However as it looks okay in most cases I think many will do sizable shifts with it anyway.

This is obviously about shoehorning in a digital back that was not designed for these lenses, and we are pushed to compromise one of the classic key features of medium format -- superior tonality -- in the red zones it becomes probably worse than your mobile phone camera in that aspect, but the center is good and the CMOS feature set is of course hard to resist...

Thanks for the analysis of crosstalk! It's consistent with my mazing artifact test results. I will upload my files to dropbox and if you are interested to take a look you are welcome. (But I have an ADSL connection and the uploading speed is a pain)

[dchew]

Thank you Anders that is very helpful. Obviously anyone thinking about this back will do their own tests, but all this work you guys are doing helps to determine whether it is worth even considering. My widest lens is the 40hr, and I don't stitch that lens. I do shift it for geometry corrections, and of course use tilt. I often stitch my others (60xl, 90hrsw, sk150) but those are less of a concern. So for me it may be worth looking into.

Again thank you CI, Doug, Anders, Bart, Yunli and others.

[torger]

Added the SK60XL to the measurements. Note that as the SK60XL is a symmetric lens, it puts similar stress on the sensor like the weakly retrofocus 40HR. If you use the 60XL's huge 110mm image circle to the edge you may even put more stress on the sensor than the 40HR at its 90mm IC edge.

[dchew]

I could not find on the CI web page if they used a CF on the 60xl. Does anyone know?

It looks like no.

Dave

[AreBee]

torger,

Will it be possible to see "heat map" results for the centred 23HR?

[torger]

Rob, I'll do a heat map for that too later, meanwhile here's more comparison for the SK60XL. My expectation for the center frames is that all are 100% okay, the reason being that the sensor has offset microlenses and thus works extra well for center frames.

The 80MP Dalsa suffers from the smaller pixels and get some minor issues, you can even see that it gets higher crosstalk number than the IQ3100 on the 60XL (12% instead of 11%), but it's more predictable and really on the edge while 3100 reaches it's peak number further in.

That the 80MP back has some more issues than the 60MP is known, and I always thought the 60MP to be a better choice for tech wide angle. However if the 80MP is good for you today, the IQ3100 doesn't lead to that sharp increase in issues so it may work out.

EDIT: added new with the 23hr. Up to 14% crosstalk, a bit more than I expected for a center frame. Not ideal, but the hot zones are pretty small and I'm more worried with the 32hr 10mm shift than the 23hr centered.

[AreBee]

torger,

...added...the 23hr.

Thanks very much.

[vjbelle]

Paul, I'm envisioning a crazy rig... Arca MF2 / Dex focal plane shutter / IQ3 100 / 'Blad CFi/e glass.  Crazy I tell you.  Id just need an ultra-wide solution.  I'm going to test my 'Blad glass with the new back on the XF Wednesday.  No shifts, but at least I can judge sharpness of the lenses.

CB

I don't share any of your enthusiasm regarding the older V lenses.  I have most of the lenses you have except the 40 and have found them all to be substandard when compared to modern Rody's or Schneider's.  Even the smaller A7rll sensor showed signs of sub-par sharpness.  Just my observations of several years......

Victor

[torger]

I don't share any of your enthusiasm regarding the older V lenses.  I have most of the lenses you have except the 40 and have found them all to be substandard when compared to modern Rody's or Schneider's.  Even the smaller A7rll sensor showed signs of sub-par sharpness.  Just my observations of several years......

There are some examples in neighboring threads, and yes they're soft at pixel peep compared to the tech cam lenses. However they seem to behave well regarding aberrations, so it's really only that softness that's the issue. They probably make fine prints.

Actually I'd like to have that softness out of my tech cam lenses too (an image that's smooth at pixel peep never get ugly pixelation regardless of upsizing), maybe some 400 megapixels or so will give me that look

[torger]

Assuming the Rodenstock Digarons were designed with the Dalsa 6um technology in mind, which is not unlikely, their target seemed to have been to keep crosstalk below about 5% when within hard vignette (as seen in my heat maps, which should only be seen as an approximate relative crosstalk measurement, a filtered LCC shot is required for more exact absolute number). It's probably a reasonable limit.

In certain conditions you can do with very high crosstalk and still not get issues, a surface which is close to neutral in color and have little micro detail can handle high amounts. Higher micro contrast and colors more different than the color of the illuminating light will increase likelihood of issues. As it happens when you shoot the things that end up in the hot zones often have low contrast, low detail, low saturation, and you're fine.

After LCC was applied I did not see any visible problems in any of the CI shots related to crosstalk, but the hot zones were exactly in such parts of the image where it's unlikely to get issues.

It's quite rare to get into a situation with the opposite effect, where crosstalk effects is exaggerated. This makes it difficult for me to decide what level is okay to live with. Sure if I had the choice to get a 60MP CMOS which behaved more like the old Dalsa CCD rather than getting this 100MP it would be an easy choice, but now it's either go with CMOS which has these properties or get a CCD. Or sit with the CCD you already have and wait until the next generation CMOS comes along and hope things improve. We've not seen BSI yet at this size. Phase One is also good at giving you an upgrade offer you can't refuse , so it shall be interesting to see how many tech users that makes the move.

[Chris Barrett]

I have most of the lenses you have except the 40 and have found them all to be substandard when compared to modern Rody's or Schneider's. 

Fortunately, with the set I've acquired, that hasn't been my experience at all.  I don't have any great tech lenses left, but here is the 40mm CFe (left) compared to the Rodenstock 45mm Apo-Sironar Digital (right).  A7r2.




-CB