40mm? That's not even a 35mm frame.
Have you seen similar issues with other Sony sensors (D800, A7r)?
40mm? That's not even a 35mm frame.
Have you seen similar issues with other Sony sensors (D800, A7r)?
CMOS sensors have traditionally not very wide angular response at all. When manufacturers do brag about it they say like "wow we have 20 degrees unlike 10 we had before!", while the medium format CCD sensors have typically 40 degrees or so (with a few exceptions). The reason for that is as far as I understand mainly the larger pixel size (which relatively speaking puts the photodiode closer to the color filter) and a different design target, eg you rather put a light shield in there and reduce the fill factor than make a very limited critical crosstalk angle.
In the 135 systems including mirrorless you use different lens designs which are stronger retrofocus so the light comes in closer to perpendicular.
The 32HR is indeed a retrofocus lens, but noway as much as the 135 mirrorless wide angles. The DSLRs are even more retrofocus of course due to the extra flange distance dictated by the mirror box, so the D800 sensor have even less reason than the A7r to have a high critical crosstalk angle.
Probably it's not too hard for Sony to reduce the crosstalk with the current sensor, just put up some walls between pixels, it will however reduce the fill factor a bit and lower quantum efficiency, ie you get a lower base ISO and worse ISO performance. For a tech cam that would be no problem. But this sensor is not designed for that, simple as that, and then it's smarter to skip those walls and just use lenses which delivers the light closer to perpendicular. On the DF+ there's no issue.
In other words I don't think this CMOS sensor is much worse than any other typical CMOS sensor in production, a high critical crosstalk angle has simply not been a design target so far.
The disadvantage with retrofocus design is that it's harder to make sharp lenses with large image circles with low distortion. Ie, it's a disadvantage for tech camera needs. However, an advantage of retrofocus design is that you can more easily have wider aperture and less vignetting, great for high ISO shooting, so the retrofocus design found in 135 systems is not only to adapt for sensors, it's to adapt for the needs found in the all-around 135 format. So for all CMOS applications so far there's been no value in compromising fill factor to increase critical crosstalk angle.
New CMOS designs are coming however which will put the photodiode closer to the color filter, so you get a higher critical crosstalk angle (and less color cast) "for free". Hopefully these type of designs will find it's way to medium format sooner rather than later.
Below an example from Fuji/Panasonic new vs old design which was presented in July 2013:
note that they brag about improvement to 60 degrees here, but using the way to measure I've done (ie measure from perpendicular) it means only 30 degrees, ie less than say a P45+ have. This is for a very small pixel size though as new CMOS technology comes first to mobile phones and compacts, with a larger pixel size the angle would increase.
Sony have their own variants, backside illumination and stacked CMOS which produce similar results. None of these technologies have yet come to a 135 full-frame though, and it needs to come there before it can come to MF.
Arca-Swiss should release a Sony bellows for their Arca-Swiss MF-two. Then someone could test the A7r sensor with Rodenstock wides, but as far as I know noone has yet presented any such results. I assume that the A7r will show similar problems when shifted as I would guess the pixel design is the same.
Last edited by torger; 19th February 2014 at 01:49.
2 Member(s) liked this post
Here is a comparison of torger's crop of the composite *JPG* (in which I made some effort to match white balance on each frame, but was doing several hundred files of processing and was not perfect at - it's also, to be fair, hard to white balance stitched tech camera shots).
In his crop of the JPG the color sure looks pretty bad. With proper white balance applied not so much (though to be clear - absolutely there is a difference in color and some additional work to achieve white balance and localized good color - download the raws and play with them yourself).
Again - we both agree there are issues with color fidelity when you excessively shift the IQ250. We just disagree where it is a practical hinderance to photographic use, vs just a scientific curiosity. We also agree that anyone looking to make the investment in a DB for a tech camera should be aware of these issues (hence the entire reason for our spending several days testing and making the raw files available) - though I also think the end user should do their own testing (I realize from your point of view in a country/market where it's very hard to do such testing this seems like a large ask - bear in mind for most of our customers here in the US it's a phone call away from arranging).
I have to leave it at that. I have other tests to process, evaluate, and share.
Not a big deal to change wb to match the colors in the the corners...
But how do the centers of these shots look, do they match in color also?
Move on to the next one.
It's a shame that Doug has moved onto other tests, because this would appear to be a pretty critical one to evaluate properly for those customers who consider colour accuracy to be of critical importance. Hopefully someone else here can spare the time to look into it to validate (or otherwise) Doug's conclusions that this is just a scientific curiosity.
(From torger's analysis, we are not talking about "excessive" shifts here. We're talking about issues that would occur without any shifting whatsoever.)
1 Member(s) liked this post
The greens still suffer quite a bit even after Doug's rework (much appreciated) on the 250 example green dress.
Reds and yellows look good.
Hopefully someone will be able to test the 250 in a outdoor environment when spring gets here and all the leaves are green again with some moderate shifts.
It is pretty obvious that the IQ250 is not ideal for tech wide angle lenses specially when stitching for critical use but produces good quality results. If you want high quality results with tech wides the best all around back seems to be the IQ260 and then the IQ160 or 180 or 280.
That said the IQ250 should work great with an fps type unit and a SLR lens like the 24mm TSE-II and of course superb with any of the MF SLR platforms.
Short story: Doug and I have different views on how a high end system like this should be evaluated and recommended, and therefore we end up with vastly different results. The reader is free to make up their own minds of which approach that's better. I still prefer my approach, but I bet some agree with Doug too and will be happy using the IQ250 with tech wides despite its liminations.
My approach to this is that MF users in general is crazy about all aspects of image quality and that includes color fidelity. See how much D800 gets bashed for its inferior color. That IQ250 then should have some free pass to fall short in color fidelity while the DSLRs don't get any slack at all I feel is very biased way to look at it, and that's not my way of doing it, my loyalty is with the buyers of photographic equipment. Another advantage of MF that is very often emphasized is the less amount of color post-processing requried, which also is an aspect to take into account.
Note that my analysis of the IQ250 files is an ongoing process, at some point I will try to produce the best results I can from these files, but I have not yet come to this step in the process. My current step in the process is analyzing how the sensor performs, and here I see a clear crosstalk problem, problems which are not seen with the traditionally recommended CCD+tech wide combinations. I think that is important to note. I get the sense that because this IQ250 has live view and that is great for tech cams, we're so excited by that so we're kind of letting go the very high quality standards we had before. Noone was recommending the P30+ with tech wides, as there were much better alternatives.
With selective color adjustments crosstalk problems can be worked around to some extent, but I still feel that it's highly unwise to build a tech cam system with lenses that produce crosstalk on the sensor, and I would not recommend such a system to a high end user, even if you can with post-processing hacks equalize the color. With post-processing hacks like this you reduce the the tonality performance, and with selective needs you end up with hours in color-tuning in post-processing, and what MF is supposed to do is minimize that kind of work. If you stitch and need to apply a different white balance and tonality settings on the different segments to make them match, that's not what I call a well-behaved system.
There's one thing if you mainly use the back for your DF+ and occassionally use it on a tech camera, and an entirely different thing if you're a professional interior photographer and intend to use the back for wide angle photography in production. I'm thinking about the latter.
As far as I can see Doug and I have a different view on how a system should be recommended. In my process scientific measurements is an important component. The thing with gradual degradations of performance and subject-dependent and even orientation-dependent issues, like the crosstalk phenomenom is, is that it's easy to get into situations when it works kind of fine for one subject, and you get much worse performance in another situation. Things like reduced tonality performance is also hard to detect, but you would not be thrilled if you knew your $35K back got its tonality reduced to Canon-like performance .
Therefore I think it's good to support the field testing with scientific measurements that makes a clear definition where this system provides stable performance. As far as I've come with this I can say that the crosstalk-free and thus stable performance region of the IQ250+32HR combination is an image circle of about 40mm. Sure you can get okay color performance outside with mild crosstalk, the degradation is gradual and subject-dependent and you can hide the effect with post-processing, but a sensor+lens combination where the sensor has to work beyond its critical crosstalk angle and thus get color crosstalk is broken by design, and I still maintain that those should not be combined in a high end system for production use. Anyone is free to disagree with me on that. I just want to be clear what my performance targets are.
Doug seems to me to have a more end-result-based approach to the problem. If results from a few real scenes look good to the eye after post-processing then it's fine and the system can be recommended. It's a practical "real photography" approach. I understand that Doug is a bit angry with me that I have used his raw files to present my measurement-based results (which is an ongoing process) which are much less optimistic about the IQ250+tech wides than the "real photography" test cases he has done. I'm very sorry for that. It's not my intention to make someone angry or misuse anyone's raw files or so, I just thought that my expertise in dealing with raw data would be appreciated by potential buyers in this forum.
In any case, I still don't think that the "real photography" way of testing a high end system like this is sufficient. It's surely a key aspect, but not sufficient. I've already mentioned why: with the real photography tests you can miss gradual degradation performance issues that can be easily seen with measurements and or just looking at spec sheets if we had them. I simply do not think it is okay to present a combination to the user which has crosstalk so close to the center, especially to users that like the MF backs for their superior tonality and color performance.
With good contacts in Phase One and in Rodenstock (which I don't personally have) it should hopefully be easy to find out the critical crosstalk angle of the sensor and the angular design of the Rodenstocks, so it would be easy to find out which system that work by design and which that do not. I think that when selling a $35K back with a $7K lens to the customer has a right to know these things.
I took the words in the original post for real: "Go crazy guys. Ask questions. Tear the files apart. Find the good. Find the bad. Let me know what you think of my effective image circle evaluations". That is exactly what I have done here, and I'm sorry I don't agree with Doug's conclusions, my image circle is considerably smaller, but I just can't change them on the basis of the results I have so far and the definition of how high performing MF systems should work. My main question is what the critical crosstalk angle of this sensor actually is according to Sony, and what the design target of the Digarons is in this aspect. Is that too hard to find out? I think it is a highly relevant question.
I will surely try to see how far one can go into crosstalk space and still get okay results, I'm just not there yet. But I see that as some kind of hack which you need to use if you happen to end up with a poor sensor + lens combination, not something that you should recommend to users. In general you can with smart software algorithms hide away serious performance issues of a system, you can never restore it in full but make it a lot better, I do this a lot myself, but I think the user has a right to know about something as serious as crosstalk, because then we are in guesstimate space. In a theoretical noise-free system color cast can be 100% recovered. Crosstalk cannot.
Last edited by torger; 20th February 2014 at 00:49.
8 Member(s) liked this post
I sometimes complain about MFDB manufacturers not caring enough about tech camera users. I know that too is not a popular opinion among some dealers. Ok, fine, agree to disagree.
I think this is another sign of it. If P1 had a clear interest in tech cams would do the simple thing to check the data sheet of the Sony and the data sheet of the Rodenstock and Schneiders and make those numbers available to dealers so they could be formed into good lens recommendations. Guesswork from what C1 algorithms can hide in some situations I do not think is a serious take on the problem. I'm sure many dealers won't agree on that, but I think I'm free to have that view and actually I think mine should be considered less controversial if you look at it from a buyer's perspective.
When you design a camera system you make sure sensor and lens play together in concert. The Rodenstock wides have been designed with certain sensor properties in mind, probably what is found on the the 6um CCD sensors. The Rodenstock designers have known the critical crosstalk angle of these sensors and adapted their optical design to that. Sure there's color cast, but that's reversible unlike crosstalk.
When you then sell a new back with radically different sensor properties the original design criterias can break, and that is what's happening with the IQ250. If I had been a dealer I would find it very difficult to suggest a lens+sensor combination which I knew broke the design criterias and require software algorithm guesswork to hide. As said, while colorcast is 100% reversible (with the exception of a bit DR loss of course) crosstalk is not.
The reason I write these long explaining posts of my position now is that it has come to my attention that my pessimistic view on the IQ250 with tech wides is by some seen as unfair, even to the point that some get angry. I maintain that my points are both valid and fair, but sure for someone that think live view is more worth than color fidelity fine at the high end, go ahead use it with these combinations. My recommendation is however to pick the IQ260, IQ160, Credo 60, P65+ or any other 6um Dalsa sensor technology to use with these wides. Furthermore based on my results I do not think it's a good idea to upgrade P40+ or IQ140 (44x33 6um Dalsa CCD) to IQ250 if you are a tech cam wide angle user that actually use shift or tilt.
By actually having the IQ250 back and a lens lineup I could make much better tests and come with more nuanced conclusions. I would still use measurement-based test as an important aspect of testing, and I would still not recommend to use the sensor in crosstalk mode as it compromises the fine color rendition properties. But I could make a much better assessment of exactly how bad crosstalk will hurt your colors and complicate your post-processing needs, if someone still would like to push its limits. But I would point on an image circle diagram which show "no crosstalk inside this circle" and say "outside this border, you're on your own". The supposedly superior CFA design doesn't count when colors are mixed. I think that matters to users.
As co-author of Lumariver HDR and RawTherapee where I specialize on various algorithms I know a thing or two about raw processing. I'm also myself a tech cam user which shoot landscape on a recreational basis. This makes me a bit emotional about this subject, I have quite strong ideas on what makes up a well-behaved system, and what's fair to fix/hide in post-processing and not, and what the customer has the right to know about the substantial amount of money that goes into these systems. To me it's a big difference between color cast and crosstalk. Reversible vs not reversible. And it doesn't end there, green separation which is a crosstalk side effect makes the back sensitive to particular LCC and demosaicing strategies. I would not be surprised if Lightroom makes a much worse job on this particular back than C1. I think it's a bad idea(tm) to make a high end system dependent on advanced guesstimate algorithms. Therefore I think what's the "crosstalk-free image circle" is a very central piece of information.
I have a somewhat faint memory of that there were rushed tech cam upgrades from P65+ to IQ180, and they did not all turn out too well, as the IQ180 has much more wide angle issues than the P65+ and IQ160. Did the dealers give proper information to the customers about that before upgrade was sold? It would be sad to see people rush from IQ140 to IQ250 and be semi-disappointed in the longer term when it's clear that it actually doesn't perform as well as it seemed in the first tests with the favourite wides. Presenting the actual crosstalk free image circles where performance can be guaranteed would avoid that situation.
Last edited by torger; 20th February 2014 at 03:31.
8 Member(s) liked this post
You have nothing whatsoever to apologise for. Your posts on the adverse effect of crosstalk on the IQ250 have been presented in an unbiased way. No fair minded reader - especially those of us familiar with your posting history on GetDPI and LuLa - could arrive at any conclusion other than that.
As you rightly point out, Doug invited others to "Go crazy guys. Ask questions. Tear the files apart. Find the good. Find the bad. Let me know what you think of my effective image circle evaluations". Therefore, Doug has no grounds on which to complain, though he understandably can be forgiven for not appreciating to learn of bad news.
There is nothing wrong in principle with a real-world point of view. However, to hold a real-world point of view is to accept an end result despite other effects, adverse or otherwise, which may exist. A person ignorant of the facts is unable to hold a real-world point of view.
Far from being criticised for reporting about crosstalk, in my view you are to be commended. For my part, many thanks for bringing to my attention what I agree is an important issue.
www.robbuckle.co.uk7 Member(s) liked this post
FWIW, your long and detailed posts are extremely interesting to read, even from a purely educational perspective. My dealer approached me wondering whether I'd be interested in the IQ250. My reaction was "nope, purely down to the sensor crop". However, subsequent to Doug's files being made available, I did start to backtrack on that a little. In some respects, the IQ250 looks very good indeed.
This cross-talk issue though is clearly a fundamental design flaw (as viewed from the perspective of a tech cam user). I was never even aware of the concept before you highlighted it, and I suspect nor were many others.
I'm sure I'm not the only one who is very grateful you have made the effort, and hopefully there will be some recognition of this issue publicly by "Team Phase One".
Simply put, if you shoot with wide angle tech lenses and colour accuracy is important to you, it would appear to be a no-brainer. Steer clear of the IQ250 - there are plenty of other great backs from Phase One to consider instead.
4 Member(s) liked this post
"Creativity takes courage." ~ Henri Matisse
Darlene Almeda, photoscapes.com4 Member(s) liked this post
I'm not "angry".
I just wish you had the opportunity to do some of your own shooting with this back and make evaluations based on that experience and not solely on scientific evaluation of the raws. I'm sure you'd want the same thing if that was practical. But I realize that's hard given your location. If you're ever in NY...
Still I posted the raw files for the exact purpose of people examining them and posting their thoughts both good and bad, so I cannot complain. We'll just have to agree to disagree the exact size of the usable image circle for these various lenses for practical photographic applications.
I did reexamine all the files last night and lowered the sizes I listed by 5-10mm on a few of the lenses for the 250. Can't hurt to take a more conservative view.
Guys I really do not know what the big deal is. I mean, it is not like Phase One is only making the IQ250. The other IQ backs are still available and are superb.
The IQ250 is the best back for available light work with an SLR in almost any available light condition. Simple as that. No other digital camera offers the same combination of low light performance, high resolution (50 MP!), tethering stability / capability (WiFi also) and Color Characteristics as the IQ250.
Some people micro focus on what the back does not do best and forget about what it can do amazingly well. Sometimes its best to pull back a bit and see the big picture.
1 Member(s) liked this post
1 Member(s) liked this post
In essence then, if you took a vertical image and then reversed the back 180 degrees you could observe the phenomona in a sort of mirror like comparison on the same subject.
Might be interesting to do this some time and see how ones sensor performs and if one orientation is 'better',with one position or the other, and also what type of subject effects the color in what manner. A user might discover that vertical works better in one orientation, with a particular subject better than the other. In other words, turning left to vertical rather than right, or vise-versa etc...
Thanks for all your work Torger and Doug.
Last edited by RodK; 21st February 2014 at 13:51.
US Representative, Arca-Swiss International
R-Line Technical Cameras, Large Format View Cameras, Ballheads D4, D4m, P1, P0, Z1, Z2, C1 Cube.
You can get crosstalk with "tech cam friendly" CCDs with the SK35 which is a pretty extreme optical design in terms of angle. I have the SK35 and I do start to get crosstalk with my 33 megapixel back, but need to get quite far out (farther out than I want to use it because of its sharpness falloff so I don't see it as a problem). I did a quick color checker test and colors were affected as one would expect, some colors more than others. I might do a more detailed test and present some data at some point, info of how to differ between color cast and crosstalk etc.
I may have over-reacted a bit on offline feedback I got on my testing. In any case I just try to be as fair and correct as possible, and just want to raise a warning so people don't rush for an upgrade without knowing what compromise is being made. I think few knew about the difference between colorcast (reversible with LCC) and crosstalk (not reversible with LCC), and that it for this particular back is an important thing to be aware about.
If your lens and shift combination is not pushing the sensor into crosstalk, all will be fine and orientation should not make any difference. If your sensor does crosstalk, its orientation may affect the result, I do think it will for the IQ250, but it's hard to predict. The only thing that happens is that the crosstalk flow changes a bit, you change from one incorrect result to another incorrect result, one might look better for the particular scene shot, but surely one would not like to have this in a production system.
An interesting experiment would be to test well-established combinations like the IQ260 + 32HR and see if there is any color fidelity falloff. The measurements I've made are quite crude, ie I need considerable crosstalk to register it, and I register it through the side effect of green channel separation. On the IQ250 crosstalk was clear quite early on as discussed. When looking at the IQ260 I saw no crosstalk whatsoever, but possibly there could be a small amount anyway towards the edge. As the quality reduction is gradual it's thinkable that lens designers accept some minor crosstalk towards the edge of the image circle assuming it would not hurt color too much. Possibly there are optical effects that could affect color fidelity too if angle gets too high (coatings?), I don't know. Therefore I think the experiment would be interesting also for a well-established combo to get a baseline of how good it can get.
Does the IQ250 Sony sensor have the same type of microlens offset the Sony in the A7r has? See image:
The results I get indicate quite strongly suggests that this is the case.
For example, the painting with the dress used in a crop further back in the thread have very different color in the shots 15mm shift up vs the 30mm shift up. The 15-up shot has considerably less crosstalk and thus looks better. The difference is that the painting is registered in different places on the sensor.
In the bad case (30-up) the painting is registered on the lower part of the sensor, where microlens offset would expect to see a low part of the lens projection, but instead it gets a high part and thus the light comes in at the opposite angle: worst case!.
In the good case (15-up) the painting is registered on the upper part of the sensor, where microlens offset and lens image projection work in the same direction: best case!.
This microlens offset makes the sensor hard to work with in shifted position, and explain why it works okay when located in the center of the image circle, but how drastic degradation you can get on one side of the sensor (and less so on the other) when shifted.
2 Member(s) liked this post
Although there could be microlens offset as described in the previous post, there could be just regular non-uniformity of pixels, ie that the amount crosstalk for a certain larger-than-designed-for angle vary a bit over the sensor. The much worse top left corner of the 30 up is hard to explain otherwise. It could be a combination of both.
I have attached here small size "scientific" renderings of the 15mm up and 30mm up IQ250 32HR raws. They have been both LCC corrected, but reference color of the LCC was from the center shot, this to avoid slight whitebalance differences that otherwise could occur.
White balance setting is exactly the same for both shots, as well as the contrast and saturation parameter. The exposure has been adjusted manually to approximately equal, as LCC correction (vignetting etc) shifts the exposure, and I'd guess copal shutters have been used which vary too a bit. To make images brighter I've clipped the windows (would have to use tonemapping to bring them in, the DR in these images is impressive). Software used: RawTherapee, Lumariver HDR for the LCC (not yet released version), and Adobe's newly released IQ250 dcp profile (which is 2.5D, ie no exposure-related color corrections to worry about).
The purpose of these renderings is to show how the sensor generates colors before you've started with your Capture One and/or Photoshop magic to try to match them. If there had been no crosstalk the color would look the same on both images. Crosstalk leads to desaturation and/or casts which is still there after you have applied LCC.
The green rectangle shows the painting discussed in the previous post which could indicate that the sensor has microlens offset. The blueish top left corner of the 30-up (worst area) does not really match this (in that area lens and microlens offset should be in the same direction), but I think the angle there is too far into crosstalk to produce any sane result, ie it does not necessarily contradict the existance of microlens offset. If you look at the topmost edge of the 15-up picture it's much better color fidelity there than in the corresponding area in the 30-up (note especially the painting with the red dress in the top-left corner of the 15-up, huge difference!). Also this is a sign of that the sensor could have microlens offset.
So what does this say? It says that if you shift the sensor too much the whole surface suffers, like in the 30-up shot. In the green rectangle I'd think the color fidelity falloff is too large in the 30-up, but may be acceptable in the 15-up. It could be the case that a 20-up shot would still produce acceptable results, but say at 25 you start to see a sharp falloff. The exact capability of the back is hard to specify without very detailed testing. If you want to be on the safe side and not have any crosstalk at all, which is what I would recommend, the 15-up shift is already too much (the upper half of the image has some crosstalk). If you can accept crosstalk if color fidelity falloff is not too bad, there's a whole lot of "it depends". You can't specify an image circle where it works, but you must specify sensor positions that work.
Last edited by torger; 24th February 2014 at 10:16.
From looking at the image in your post, it appears the the entire upper left suffers, much past 15mm. I base this on the blue hue?
I am still wondering where the horizontal shift falloff is, I am assuming somewhere between 7mm and 10mm of horizontal shift?
That's a pretty significant difference in colour. I think Doug's post above (three images of same crop with green dress) is (unintentionally, of course,) misleading with regards to overall colour rendition and fidelity when employing movements. For example, if you correct for just that part of the image, the rest of the image suffers and visa versa. It seems to me that if you own a technical camera, the IQ2250 falls short of being an idea back. Bring on the next iteration!
On the other hand if Doug had skipped the 30mm-up shot (where color breaks apart completely) and only stitched the 15mm-up the color problems would have been much much less. Haven't detailed-studied the 15-up-15-left/right yet though could be significant problems in the corner, but need to check that.
If you look relative to the sensor size 30mm is a very large shift for a 44x33mm. In my landscape shooting style I would rarely do more than 10mm shift on a lens this wide with a 44x33mm. For interiors I'd probably need a bit more though, and if you stitch to simulate a larger sensor area you'd need more too.
Since I got aware of this crosstalk issue I've noted that the old SK35XL introduce crosstalk on CCDs, even my old Aptus 75, and probably it's a bit worse on a P65+ or similar 6um back. The effect is desaturation and slight hue shifts towards the sides, can't get as drastic effects like with the IQ250 though (probably due to that the sensor's critical crosstalk angle is not violated as much), but the point is that people have made pictures before with some crosstalk in them and been satisfied with the results. Many sold their SK35XL though with the P65+ and other 6um sensors, most complained about lack of sharpness rather than lack of saturation and color accuracy though, probably because we did not know that one should look for that.
If I had the opportunity to in person test the IQ250 with some lenses I'd do it and try to find more well-specified limits, say 1) crosstalk free image circle, and 2) maximum shifts with crosstalk no worse than it was for a SK35XL and a P45+ which was a high-end configuration of the past whose preformance many can relate to. I think one could then end up with a usable system for moderate shifts on the wides, but you'd need to compromise color a bit, as we apparently did with the SK35XL.
If you today are used with the color performance and shift stability of a IQ260+32HR I don't think you'd want to downgrade that, but a user going from a SK35XL and P45+ to a IQ250+32HR could become satisified (note I haven't tested the P45+/SK35XL in person, there's a slight risk that my Dalsa Aptus 75 has more crosstalk than the Kodak in P45+, but I don't think so).
I've attached an image that illustrates the desaturation, by using the IQ260 image as reference. Decide for yourself if this is acceptable as a starting point. As said, with localized edits in photoshop you could match the colors a bit better.
(The IQ260 and IQ250 were white balanced with a color picker on the same spot, but profiles differ a bit and there can be crosstalk color shifts too on the IQ250 so the color of the blueish green dress is different. The purpose of this image is however to see how little the pale yellow background differs on the IQ260 image, and how it has been desaturated due to crosstalk with the IQ250)
2 Member(s) liked this post
Wow, thanks. It's really interesting to see this stuff and makes me realise how insanely hard it must be to design, build and implement these sensors and related technology. Again, this shows a big difference in colour. The IQ260 is still king in my book, but that's because I'm a tech camera shooter. For others, I'm sure it's a God send.
Here's a trick if you want to test your lens+sensor combination for crosstalk that is a little simpler and quicker than shooting a colorchecker in various positions and comparing:
1) shoot an LCC with white light (ie a normal LCC shot, ambient light is generally okay)
2) shoot an LCC with magenta light (or use a magenta filter on the lens, or a magenta LCC card if you can find one), the purpose is to strongly reduce the green channel content.
3) correct the magenta LCC-shot with the white LCC-shot
4) you will now have a flat magenta image, white balance it in a place that is as close to the center of the lens as possible (where we assume there is no crosstalk).
5) Result: if the whole image now is a flat gray without casts, you have no crosstalk, if you have a residual cast (should be green) you have crosstalk, the stronger color the more crosstalk.
The reason the cast is green is because we made a magenta shot (very little green content) and light filtered through blue and/or red color filters on the sensor has leaked into the green photodiodes.
The attached image shows the result with an SK35XL and Aptus 75 with extreme shift, which has some crosstalk issues. The crosstalk is much stronger for horizontal shift than vertical (with sensor in landscape orientation like here), as the sensor has tungsten bars horizontally that reduces crosstalk for vertical shifts.
I think this test could be a good one to do in addition to regular color cast tests when testing a new sensor on tech wides. Note: I've used Lumariver HDR for this, haven't tested if the test procedure works in Capture One (or Lightroom) but I think it should.
While there's no mathematical solution to make a true inversion of crosstalk, it might be possible to make a compensation which in practice yields a good improvement, ie fix automatically what you would otherwise have to tweak manually in photoshop or leave as-is. I might make an attempt for that. The workflow would probably require two LCC shots like in this example. I'm not sure yet though if there's enough information in the magenta shot to make a decent compensation. If you make shots for red and blue light too you have even more information to feed an algorithm, but at some point the workflow would become too complicated still without possibility to make a true inversion.
I'll keep you posted if I make progress with such an algorithm (would be included in Lumariver HDR). It can help out the IQ250, and bring better colors to your SK35 shots.
Last edited by torger; 28th February 2014 at 00:45.
2 Member(s) liked this post
Added some preliminary results from crosstalk cancellation algorithm development here:
With some luck it may turn out useful for the IQ250, it's early in the process though.
Thought I'd resurrect this thread and ask if anyone out there has bought and is using the IQ250 with a tech camera and the 32HR in the field? I haven't seen any test images that show how all the above plays out in the real world where bright blue skies etc. might make the crosstalk and colour fidelity issues very apparent. Any more info of test images would be greatly appreciated.
I have an IQ250 and the 32HR. If you can let me know exactly what you're after, I'll try to take some shots in a few days (so far I've only used it unshifted on the FPS, but I have a Max so can use that).
With the IQ250 I think the issue is the proximity of the of sensor to the rear element of the lens in use with a tech camera, in that the close the rear element is to the sensor, the more issues with color sat loss on movements will occur.
I can provide a bit of feedback, from tech camera lens standpoint. I recently was able to test the Arca M2, setup for a DSLR mount, sold as the DSLR2. With this setup, you can mount a Sony A7r and use almost all of the current lenses from the tech lineup, exceptions, (23 Rod, 28 Rod). The wider the lens, the closer it has to come to the sensor, so I expected good results with the 60XL and 90mm Rodenstock and OK results with the 40mm Rod and 35XL.
I realize the A7r is not the same size, or MP as the Sony sensor in the MF cameras now on the market, however I believe the Sony 35mm sensor and the MF cousin have similar characteristics. The Sony 50MP sensor in the various MF cameras out there, Phase, Hasselblad, Pentax is the same base sensor, and it uses micro lenses. Extreme shifts on these types of sensors, does seem to have issues. I don't know if the individual pixels on the 35mm Sony are larger or smaller than the pixels on the 50MP version, it's my thought that they would be larger.
What you are going to see with shifts up to 10mm, is excellent to good results, dependent on scene. As you approach 10mm, you are going to start seeing more color/sat fall off, no loss of details. On shifts up to 15mm you start to see significant color/sat fall off, as much as I feel 40% maybe even 50% depending on the lens used. Past 15mm and things really fall apart. You hit the edge of the sensor by around 18mm anyway, so pushing to 20mm was moot.
Of the lenses I used, I found the following: LCC's were shot with each.
35XL, excellent on center, good to 5mm shift, OK to 8mm of shift, very harsh color loss past 8mm and strong magenta cast.
40mm Rod, excellent on center, good to 10mm shift, Very harsh color/sat loss by 15mm
60XL excellent on center, very good to 10mm, good to 15mm, useable at 18mm but some work required on color recovery. CF was used on 60XL
90mm Rod, excellent all the way to 18mm.
You can assume the 50MP CMOS in the Phase IQ250, will have similar results with tech lenses. Maybe a bit better but the loss of color/sat on shifts of 12mm or more is pretty considerable. The 32mm Rod may do a bit better, but when you look at the total cost of that lens I can't see the justification if you are going to be limited to 12mm or so of shift.
Is recovery possible with LCC?
Yes and No.
Yes, in that I found most times in C1, I could recover the loss of green saturation and hue on extreme shifts, not perfectly, but enough that I could make a print.
No, in that a blue sky is very hard to balance out as the loss of blue hue/ color shifts is the hardest to recover. This is especially true if the sky is pure blue with no clouds to help break it up.
The amount of time in post to recover the files to me was extreme mainly due to the color/sat loss on shifts.
What Gerald is seeing with his IQ250 and the TS-E lenses is totally different as with these lenses, the lens is farther back from the sensor (as I understand the setup), thus the color sat loss issues will be much less extreme to none. The 24 and 17 TS-E's are designed to work with a mirror box which should allow the lens to be much further from the sensor at infinity.
Just how close you have to come with the 35XL is around a 1/8" gap, and the 40mm Rod is around 1/4". The 60XL allows for a much healthy gap, approaching around 1/2".
Thanks for your comments, Gerald and Paul.
I guess I'm hoping that this Sony sensor is somewhat of a magic bullet, particularly when I think of it in the context of Hasselblad's new CFV-50c and Nick-T's assertion that live view can be enabled on it via a Firmware update (I'd NEVER buy on rumors though, just to make that clear). I think the IQ250 represents the gold standard when considering interface, internal battery and software, but it's hard to argue with the price difference between the two products.
Anyway, I'm hoping that this sensor might give me all the movements and versatility I require when mounting on my Linhof Techno. My first look at the DT test comparisons led me to believe that the IQ250 pretty much trounced the IQ260 in every aspect of image quality. Subsequent reading (mainly of this thread) and looking at the files suggest otherwise, however.
For my purposes I need to be able to use a 25/28mm equiv. lens (35mm terms) and be able to employ an absolute maximum of 18 to 20mm rise / fall. I'd also need to be able to employ 2 or 3 degrees of tilt in conjunction with up to 10mm rise / fall. I am a documentary and fine art photographer and mostly shoot landscapes and urban scenes, with the occasional environmental portrait. Colour is very important to me, which is why I've stuck with shooting film and scanning on an Imacon 949 for so long. I guess this new generation of CMOS DMF backs has me exciting because the exposure latitude / DR seems amazing, the on back live view will do away with the need for using a sliding back adapter (I don't stitch) and, in terms of the Hasselblad at least (providing they do implement LV) the price is becoming more attractive.
Anyway, I was just wondering if anyone out there is using an IQ250 on a tech camera and what they think about the performance within the technical requirements I've listed above. I'd be very worried about colour fidelity of things like blue skies, green grass, etc. There's no point investing this kind of money in something that only gives you a portion of what you need. I'm guessing people are going to say that for what I want the IQ260 is still the gold standard, but I live in hope.
The work you did with that set up was very enlightening; thank you for the effort. One point about the proximity of the rear element. In general that's correct - the closer the element is the worse the color issues. However, it is really the angle of incidence at which the rays hit the sensor. For example, one could imagine a lens that has a rear element only mm's from the sensor, but if that element was huge, and hence projected light rays out of the lens perpendicular to the sensor, there would be no cast at all. There may of course be other serious problems with such a design!
I bring this up because I am about to receive a 60xl. I am very curious how the 60xl does vs. the 70hr, and relative to the 40hr. the 60 fills the gaps in my lenses a bit better than the 70, and I also hope it to be a great single-lens choice when going really light. I'm sure its cast is worse than the 70, but I wonder by how much. Your report shows it is actually better than the 40hr, which is encouraging. I have the IQ180, so your results won't correlate directly with mine, but encouraging non the less.
Once I get the lens, I should also be able to borrow a 250 from my dealer. I could do some "quick" (ha!) comparisons of these three lenses sometime in the next few weeks.
The only solution I see going forward with WA lenses / tech cameras and CMOS DBs is a new generation of (even) stronger retro-focus lenses.
Just can't see P1/Leaf, HB, Leica, Pentax or Sinar getting Sony to make a chip that behaves well with large movements using the current crop of WA lenses. Sony probably only wants to fabricate one MF chip and then sell it to as many OEMs as possible; Leica and Pentax have no need for a chip that behaves well with large movements for a start, so that puts even less pressure on Sony to alter the design. Most DB users probably use their DB on an SLR-style body, so they'll be happy with the current low noise/high ISO CMOS implementation.
In the short term, I think the most flexible tech-cam solution is something like a P45+ for large shifts, with an IQ180 for single high-res shots. Both together should cost less than a used IQ260. If small movements are required and you have R/S WAs, then the CFV-50c is an attractive proposition. If no movements are required - e.g. you shoot with an Alpa TC - then the CFV-50c is a no-brainer. I'd rather have the low noise / high ISO / long exposure capability / $20k saving of the CFV-50c than the extra 10MP of the closest competing CCD back, the IQ260.
1 Member(s) liked this post
Thanks. Yes, I think I've been lost in some very wishful thinking regarding the CFV50c / CMOS MFDB. If I'm being practical, I think I'll likely end up with an IQ160 and have to suck it up and use a sliding back.
I use the back on a Arca Swiss RM3Di with Rodenstock 40mm and 70mm HR-W lenses.
I love the setup works well for everything except night photography and extreme high iso but from dawn till dusk it is my favourite rig for landscapes.
The IQ250, from what I have seen posted so far, seems to work superbly with SLR lenses so the best choice for wide angle work with that back would be either one of the SLR systems (Hasselblad and PhaseOne/Mamiya) that have good wide angle lenses available but if you want tilt and shift with extreme wides the only choice would be an Alpa FPS system and Canon TS-E lenses. (Hartblei H1 might work as well). Of course the IQ250 works very well with longer tech camera lenses but that is not the issue.
The Benefit of the IQ160 and 260 backs is that they work well with almost any lens made today all the way down to the 23mm Rodenstock which on a sensor that large is equivalent to a 15mm lens on a 35mm full frame format DSLR.
1 Member(s) liked this post
If you have time, I'd very much appreciate seeing some examples of the 32HR shifted on the IQ250. Any examples would be handy, really. I'm curious about performance more than anything, so not urgent, but I'm sure others would be interested too.
1 Member(s) liked this post
No problem, whenever suits you. Have a great holiday!