Phase One IQ250 Tech Cam Testing (vs IQ260 vs IQ280) by Digital Transitions

[Doug Peterson]

The IQ250 files can be processed in RawTherapee by the way, but probably there need to be some matrix fixing before getting full support.

Iridient Raw Developer can also open IQ250 files as of their latest version. They also natively support EIP files, so no need to extract them. It seems to handle the Mazing in some of the extreme combinations (e.g. 30mm rise and 15mm shift with the 60XL on IQ250) better even than C1.

[Doug Peterson]

It doesn't seem to be as bad as I expected.

Was a center filter used on the 32mm?

A Center Filter was NOT used.  I'll add some notes to the test page to that effect.

I'm also adding a statement in our summary that users that previously did not use center filters might want to consider them in light of the larger amount of light fall off we see in the IQ250, but that this might be offset by the increased ability to recover shadows in the IQ250.

[lelouarn]

Doug, thanks for doing this incredible amount of work !

I had a couple of questions:

- How did you manage to keep track of all the settings for ach picture through the shoot ? Since there is only partial exif on tech cams, you must have had a way to assign the aperture / movements / lens used to each image. Was it simply a notebook / excel with notes associated to each file (filename displayed on the back) ? Or did you shoot tethered, and then filled in the exif after each shot ? Or something more clever ?
I'd like to do some tests on my own, but keeping track of things is a bit tricky.

- You say on DT web-page that you shot the LCCs later. This means that setting the aperture back to f/9 is repeatable enough ? There are not click stops for the aperture on the shutter, so I've been a bit suspicious of doing the LCC later (as you can not be 100% certain you come back exactly to the same aperture - could be 9.2 or 8..

Thanks again for this testing !

[Doug Peterson]

Doug, you wrote:

"Processing/Color: Processing of the composite files was done in Capture One v7. White balance, which is hard to determine in a mixed-light moody scene like this was set roughly during processing to attempt to match the several backs and lenses in the same ballpark - however, this is an imperfect process especially as some of the lens+back combinations had uncorrectable color cast. We do not recommend this test be used to evaluate the subtle differences in color between these three backs (we will have more testing specific to color coming soon)."

Which lens/back combinations had in correctable color casts?

What are your recommendation for the max shift with the IQ250/40mm Rodenstock combination?

35XL and IQ280 and IQ250 was the worst offender regarding uncorrectable color casts. In these combinations the LCC routine was not able to restore high fidelity color to the effected areas.

This is not a new finding, and is one of the reasons we started recommending Rodenstock wides on a near exclusive basis for the IQ180.

The table of max usable image circles in the link can be used with our visualizer to determine max movements, which depends on orientation of the sensor and how much you care about the extreme corner of the image (e.g. if there is featureless sky or featureless white ceiling in the extreme corner you may be okay with several more mm of movement). In the case of the IQ250 and 40HR I found acceptable results with a 90mm IC (the same as stated by the manufacturer) which according to our visualizer means between 20mm and 25mm of rise could be used on a horizontal frame.

[Doug Peterson]

- How did you manage to keep track of all the settings for ach picture through the shoot ? Since there is only partial exif on tech cams, you must have had a way to assign the aperture / movements / lens used to each image. Was it simply a notebook / excel with notes associated to each file (filename displayed on the back) ? Or did you shoot tethered, and then filled in the exif after each shot ? Or something more clever ? I'd like to do some tests on my own, but keeping track of things is a bit tricky.

Shot tethered and used the Next Capture Naming to keep track. I made one mistake - the 60XL on one of the backs I did [top-right > top-center > top-right] meaning I missed the [30mm rise 15mm left] combo. Still kicking myself about that.

- You say on DT web-page that you shot the LCCs later. This means that setting the aperture back to f/9 is repeatable enough ? There are not click stops for the aperture on the shutter, so I've been a bit suspicious of doing the LCC later (as you can not be 100% certain you come back exactly to the same aperture - could be 9.2 or 8..

The Schneiders have clickstops, the Rodenstocks do not.

In other testing and past experience an LCC is not as sensitive as you might think. Being off by a mm of movement, or a fraction of a stop, or a small change in focus is not of real-world consequence. Obviously if you're off by several mm of movement, a large change in aperture or focus does matter. Also I wanted this to be as real-world as possible; whatever quality of results I could get in this test I fully expect photographers to get at least that good of results (or better) when faced with real world challenges. In the real world photographers are careful about setting aperture and movement the same for the LCC, but not scientifically precise about doing so.

[torger]

It seems to handle the Mazing in some of the extreme combinations (e.g. 30mm rise and 15mm shift with the 60XL on IQ250) better even than C1.

I shall try to figure out the reason why mazing occurs (edit: found the reason, read below).

With RawTherapee one can select different demosaicers and it depends on type how bad it becomes, the types that try to extract the most details get most issues with mazing. It becomes so bad it's unusable, even colors fail, see attached image. While a VNG4 demosaicer provides a smooth result.

I'm no demosaic expert, but my layman's guess so far is that it's because there different amount of pixel vignetting on the greens, ie there's some difference between G1 and G2 and many demosaicers expect that they should have the same amplification. VNG4 is an algorithm that assumes G1 and G2 are different, but can't extract as much fine detail as many of the others.

Edit: looking at the raw values I can confirm that G1 and G2 channels start to differ, and the difference increases further out. The example crop is 30mm out and there G1 is about 1/3 stop weaker than G2. Without shift the G1 and G2 has the same amplification and the image also demosaics well. That G1 and G2 vignettes differently makes it more complicated to do flatfield correction (=apply LCC), but Capture One has obviously an algorithm that can handle it quite well. It shall be interesting to test in my own Lumariver HDR but I need to fix some stuff before IQ250 files can be opened there.

RawTherapee's current flatfield correction algorithm can't handle this situation at all, while it works well for the CCDs I've tested (I use it all the time with my Aptus 75), so I'm assuming the CCDs vignette more evenly on G1 and G2.

Note to bayer array newcomers if you're confused with my notion of Green1 and Green2: there are two greens for each red and blue pixel, usually both greens have the same amplification, but in some cases they can differ. With this sensor it seems to be that amplification indeed is the same, but angular response is different causing variations over the sensor when wide angle lenses are shifted. The pattern of this variation would be interesting to investigate. Hopefully they're smooth, because if there are small spotty variations in angular response it will be hard to correct well without smearing detail or losing color stability.

This sensor seems like a very interesting challenge on tech cam wides. With proper algorithms one can probably get quite good results with the wides (one can afford to lose a few stops with this supreme dynamic range), but if algorithms are not doing the best the result can be a disaster.

[Paul2660]

The table of max usable image circles in the link can be used with our visualizer to determine max movements, which depends on orientation of the sensor and how much you care about the extreme corner of the image (e.g. if there is featureless sky or featureless white ceiling in the extreme corner you may be okay with several more mm of movement). In the case of the IQ250 and 40HR I found acceptable results with a 90mm IC (the same as stated by the manufacturer) which according to our visualizer means between 20mm and 25mm of rise could be used on a horizontal frame.

I looked at the visualizer, but still can't tell on shift.  What I am wondering is with the 40mm Rodenstock, which artificially vignettes due to the design by Rodenstock at 16mm on a 260/160/180 etc. full frame.  Will the 40mm on the 250 go to say 20mm or even 25mm before hitting the artificial vignette?  Or do you suffer too much from light fall off since the 40mm has no CF.

Just curious. I know you are taxed at the moment.  My same thoughts apply to the 28mm which eve though has only a 70mm IC, it may shift to 15mm or 12mm on the 250 since it won't see the artificial vignette as fast as the full frames will.

Paul

[Doug Peterson]

I looked at the visualizer, but still can't tell on shift.  What I am wondering is with the 40mm Rodenstock, which artificially vignettes due to the design by Rodenstock at 16mm on a 260/160/180 etc. full frame.  Will the 40mm on the 250 go to say 20mm or even 25mm before hitting the artificial vignette?  Or do you suffer too much from light fall off since the 40mm has no CF.

Just curious. I know you are taxed at the moment.  My same thoughts apply to the 28mm which eve though has only a 70mm IC, it may shift to 15mm or 12mm on the 250 since it won't see the artificial vignette as fast as the full frames will.

Maybe you missed that you can click and drag the sensor to the left or right in the visualizer?

With the 40mm a horizontally mounted IQ250 you will hit a hard vignette in the two outside corners of the frame at almost exactly 20mm of shift. But at 18mm of shift you'll be entering the mush/low-fidelity area of the image circle. So I'd only use 20mm of shift when the image content and colors in the corners is not that important.

28HR I can't tell you how much movement since I haven't tested. From the specs it should be 9mm of shift before you hit the hard edge.

[torger]

I've looked more at the IQ250 32HR jpeg. Maybe I was wrong when I said it wasn't as bad as I thought.

I think the library is great to show detail in the shifted area, but it probably masks (lack of) color stability quite a bit. It seems to me that already outside the 60mm frame the LCC has difficulty in correcting color, which would be visible in a more evenly colored ceiling.

I also see just above the "15mm Rise" label the next section getting stitched in and it has severe mazing artifacts as discussed above (probably due to different G1/G2 pixel vignetting).

From that image I'd say that the 32HR usable image circle is likely to be 60mm rather than 80mm that shows in Doug's table. You can compare the IQ260 and IQ250 32HR file side by side then it becomes more obvious that the IQ250 loses color stability rather quickly.

A sanity check experiment one can do is to LCC-correct the LCC shot with itself. If that is not a perfectly uniform surface afterwards then it's just too much vignetting/color cast to be fully corrected.

Some subjects can work with some residual color cast, this library ceiling for example can swallow more than a clear blue sky, but I think a recommendation of usable image circle should be defined as zero visible color cast even if you photograph a clear blue sky or a large white wall.

[Fine_Art]

I did not experience any mazing problems with the basic raw file in RT.

Thanks Doug for providing the raw for testing.
Here is a copy converted in RT 4.0.11.79. I tried the newer .12 version, it is a buggy mess that crashes. Stay away until it is revised. .11.79 is stable.

http://farm3.staticflickr.com/2806/12463985743_f1c563964f_o.jpg

The amount of detail in 1 shot is impressive. I did not like the colors especially green, showing up on the bars. That must be particular to the No AA plus that lens.

The DR is very good. Having the daylight of the window shows that off even with a different color temp.

[torger]

Try the shifted ones, that's where the G1 vs G2 problem occurs and you get mazing. The unshifted that you have converted so nicely has no mazing issue (I've tested that too).

The high amount of false colors (greens on the bars etc) is also showing the the Capture One conversion, it's an aliasing artifact. Stopping down to f/11 had probably been a better choice for this subject, maybe even more. Some prefer uber-crisp images and manually clean up moire/false colors though, while I prefer stopping down to kill most problems with diffraction and then sharpen up.

You can try experimenting with the "false color supression steps" in the raw tab of raw therapee to reduce false colors, if you've not already done that.

I did not experience any mazing problems with the basic raw file in RT.

Thanks Doug for providing the raw for testing.
Here is a copy converted in RT 4.0.11.79. I tried the newer .12 version, it is a buggy mess that crashes. Stay away until it is revised. .11.79 is stable.

http://farm3.staticflickr.com/2806/12463985743_f1c563964f_o.jpg

The amount of detail in 1 shot is impressive. I did not like the colors especially green, showing up on the bars. That must be particular to the No AA plus that lens.

The DR is very good. Having the daylight of the window shows that off even with a different color temp.

[Ramirez]

32HR Jpgs: i'm shocked by the noise of the iq260 in the shifted areas (over 5mm) left and right. the iq250 is almost noise free, although the image appears to be half a stop shorter exposed. highlights rock on iq250 too, clear advantage in dynamic range, resolution also better (higher pixel density, ok, but the rodie does deliver). color shift is hard to tell with this setup, would be great to see a test with an overcast sky. doug, when will you do this again outdoors? ;-)

seriously, great work doug, thank you so much!
my next step will be heading out with both of them on an arca rm3di, i think there is no other way to make a buying decision.

-----------------------------------

www.silent-moment.com

[Doug Peterson]

color shift is hard to tell with this setup, would be great to see a test with an overcast sky. doug, when will you do this again outdoors? ;-)

In my experience color fidelity is better evaluated with varied and strongly colored subjects than an overcast sky. Alpa did a test with overcast sky that fails to show the color issues I expect would have been shown with a more strongly saturated subject.

We also have done 32HR with IQ250 with strong rise to shoot the Empire State Building with blue sky. I'll be looking at those tests in more detail next week.

[Doug Peterson]

The high amount of false colors (greens on the bars etc) is also showing the the Capture One conversion, it's an aliasing artifact. Stopping down to f/11 had probably been a better choice for this subject, maybe even more. Some prefer uber-crisp images and manually clean up moire/false colors though, while I prefer stopping down to kill most problems with diffraction and then sharpen up.

Selection of f/9 was largely a practical issue of trying to keep exposure times short enough to do the number of back+lens combos I wanted to get through.

Having completed the test, were I to do a paid job shooting this location I'd probably use f/11.

[Doug Peterson]

seriously, great work doug, thank you so much!
my next step will be heading out with both of them on an arca rm3di, i think there is no other way to make a buying decision.

You're welcome!

And yes, I agree, your own test is the best way. As much as I think our test provides useful information there is no replacement for your own testing.

[Paul2660]

All you have to do is look at the shifts, forget the center.  

The 32mm had no Cf, thus the lightfall off to the shifted edge is not balanced as well as the CF would allow so the LCC is working hard.

NET, the 260, which appears to be shot a ISO140 which is the base for long exposure looks pretty darn bad, excessive noise and really the shift is not useable unless you really down rez the file to around where sensor plus would get you.  On center I can get the box where the shadows are pretty harsh to look OK, but as Eric has shown the 250 is cleaner here, but the shift is where the real difference is. Just look at the large wall section to the far left.  With the 260 it's all pretty much noise and no structure.

The 260 also has more aliasing on the X frames that cover the books, way more.  It's very harsh on the segments where you are looking down them not head on but even head on they appear to be made with beads of different colored material.

The 260 file also shows lots of  (way to many IMO) stuck pixels which really surprised me.  They are evident anywhere the image is dark or in shade and show up mainly as blue.

The 250 is cleaner throughout, but I feel the transitions of the details are smoother, take a look at the beautiful inlaid vertical strips at the edges, they look much better on the 250, also look under the cases in the middle of the shot, again much cleaner.  

On the 250 shift it's got some banding, which I actually have come to expect from this type of pull with CMOS, however you can clean that up.

My net opinion, if all you wanted was a center shot, you can make either test work, however if you were after a FULL RESOLUTION composite, you just can't use the shifted parts of the 260.  I didn't look at the right shift as it should be the same.  Doug shot these at 2 minutes as I remember so to get a better shifted image on the 260, he would have had to 2x the exposure and maybe do longer, which I believe would have blown the rest of the image out.  

Since this is a Sony chip, and I guess the their processor, it's clear that they have a strong advantage in chips.  I saw similar results when I first shot the D800 and now it's the same in the MD world.  For me it's the noise and lack of use of the shift.  If you dig way way way down, you might find a spot where the color is better on the CCD from the 260, but I can't see it.  

The 250 file non OE still pulls up very well on the shift.    

To keep a 260 is now quite a bit harder a decision to make.  I am not a big user of rise and fall, but these pretty much look the same on the 260.   I will try to put some images together as Eric did later tonight.

Paul C

[Doug Peterson]

I did every IQ260+32HR file in both ISO50 and ISO140.

I screwed up the naming and two of the six files I listed were actually the ISO140 shots.

For any evaluation of noise I strongly suggest using Capture One. While several programs have support for opening the raw files from the IQ250 nobody else has spent as much time making the files singing, and nobody else uses the dark frame information which is especially important for the IQ260.

I am correcting that now...

Edit: They are now corrected.

[torger]

In my experience color fidelity is better evaluated with varied and strongly colored subjects than an overcast sky. Alpa did a test with overcast sky that fails to show the color issues I expect would have been shown with a more strongly saturated subject.

We also have done 32HR with IQ250 with strong rise to shoot the Empire State Building with blue sky. I'll be looking at those tests in more detail next week.

Have you tried applying the LCC on the LCC shot itself? It should be telling. If it doesn't even out, then it's outside the capable range for sure.

Now when I've had a long good look on mainly the 32HR IQ250 shot I think the uncorrectable color casts may be pretty severe, but is somewhat hidden by this subject. Easiest is to show the IQ260 and IQ250 32HR jpeg side by side and compare the color in the ceiling. The IQ250 desaturates and I'm quite sure it has some cast too. I would have to do more tests to be certain though. I haven't tried the LCC on LCC trick myself yet.

[Paul2660]

OK, I have calmed down, the iso 50 on the 260 is better, but the 250 still wins on the shifts hands down. (These are not using a CF)   It might be closer if I down rez the 60 to 50mp but the iso 50 image could work. 

Darn, trying to WB that subject is like a final exam.  I appreciate a bit more that the interior shooter go through on a daily basis. 

Still don't get why the 260 is showing so many stuck pixels even at iso 50, the dark frame should get those.  The 250 has none.  The 260 does have a lot of aliasing for sure.  Shows clearly on the X frames.

Beautiful building also.

Paul C.

[Richard Osbourne]

Thanks very much for this comprehensive first test, Doug. Having just bought an IQ260, I'm slightly depressed to see just how well the IQ250 is controlling noise (and the 260 not, especially shifted). Also I'm really surprised that the IQ250 appears sharper - some of the text on the sign is clearly readable but not so much on the IQ260. Will await the fuller colour tests, but on this evidence CMOS full frame can't come quick enough for me.