DT Tech Cam Test - IQ250 vs IQ260 vs IQ280

[dougpeterson]

Phase One IQ250 Tech Camera Testing

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 - they are very much a work in progress.

I'm going to try my best NOT to be online tonight. Bit burnt out - I worked all weekend and then sat on a tarmac for 3 hours yesterday - the entire time I was working on processing and working up this test.

We have a few more tests captured, including a great ISO sweep at this same location (I say great because it's a good location for an ISO sweep - lots of dark and light tones, lots of fine detail, lots of color) but it will probably be a few more days before I get to post them as we have an IQ250 Open House in NYC on Wednesday that I have to help prep.

 

[gerald.d]

Wow. Not looked at the files yet, but clearly a huge amount of work gone into this.

Thanks Doug.

 

[RodK]

Doug,
Fantastic test! Thanks for your time and dedication in finding answers to the questions posed by these new sensors.
Test looks fantastic. Take a rest. You deserve it.
Rod

 

[Williamson Images]

What a great setting! This should be fun.

Thanks Doug.

Robb

 

[tjv]

I'm just waiting for someone to yell murder and claim false results because you used an an Arca–Swiss ballhead instead of the Cube...

 

[tjv]

On a more serious note, I think the link to the IQ260 and 40HR composite file wrong. It brings up the 35XL composite?

 

[dougpeterson]

On a more serious note, I think the link to the IQ260 and 40HR composite file wrong. It brings up the 35XL composite?

Thanks for pointing this out! I've now corrected the link.

 

[tjv]

I thought people would be discussing all the details and outcomes of this to death by now… What I want to know is, as a technical camera shooter, forgetting crop factor for a moment, which sensor out of the 250 and 260 looks the most appealing for a tech cam user who already has retrofocus, Rodenstock lenses?

On LL there is discussion about the 250 having better detail and dynamic range, and the performance with shifts and after LCCs being pretty comparable to the 260 all things considered. What's the consensus here? Being a film shooter and Imacon 949 user at present, I'm not completely sure what to make of these tests. I have suspicions that the 260 would be a better option for me, but can't quite explain what I'm seeing.

NB: I will add that I can't afford either at the moment, but live in hope. I can see a second hand IQ160 in the mid term, but this gear is a little out of my league.

 

[Paul2660]

The 250 for the most part pretty much blows the 260 away in the tests shown by DT. Especially on the shifts. No CF was used and the 260 would most likely benefit from that but so would the already excellent shadow detail of the 250.

My take away:

250 works just like any of the modern Sony sensors does (in 35mm) excellent shadow range and at base iso no loss of details in the shadows.

260 shows noise in the shifts but the areas that were shifted were almost black on the raw file. The 260 just can't pull up as well as the 250 can. The 250 actually shows the fabric of the wall on the left which also looks very dark on the 250 raw.

260 has a lot more aliasing and the shifts seem to be a bit softer, but that tends to be the case on underexposed shifting. This subject was extremely challenging.

The 260 seems to have a lot more stuck pixels even after the dark frame, I was surprised by this. Even after Capture One sets it's defaults and you move the single pixel noise slider to 100% blue stuck pixels remain. This was only a 6 second exposure on the 260 at iso 50.

To me the 250 show where the future is going, and the only question is with Sony showing no new large MF chips coming (only the 54MP 35mm chip in 2015) will the continue to work on a full frame solution for MF and if so will it be 60MP or 80MP. I have to think that Dalsa is working on something, but I will be frank here, Sony set the bar 2 years ago with the D800, and they just improved it again with the 250. Dalsa will have some work in front of them to get there and as far as I know they don't have a CMOS chip in this market place, I am sure they make CMOS somewhere in their fab plants. But the trick to the Sony chip, is the ability to pull up the details from shadows.

As a owner of a 260, seeing the results of the 250, I would attempt a move immediately if the crop factor was 1:1. I really have to consider the 1:3 and now am considering upgrading to a 280 so I would have 20MP sensor plus when I need it. In my work the 260 long exposure mode is not needed that much and the 20MP sensor plus should be pretty clean.

Paul C.

 

[kdphotography]

I'm just waiting for someone to yell murder and claim false results because you used an an Arca–Swiss ballhead instead of the Cube...

Uh yeah, Doug. Could you do these all over again using the Cube?

:ROTFL:

 

[torger]

I'm a bit concerned by the uneven pixel vignetting on the green channels of the IQ250, and also how color is affected by the color cast. I'm afraid that if we instead of a colorful ceiling had an overcast sky I we would see uncorrectable color shifts with the IQ250, and we would reduce the usable image circle diameters substantially.

I think I see that already in Doug's example but it's hard to evaluate as lighting condition could change in the ceiling too. Putting the IQ250 32HR and IQ260 32HR jpeg images side by side makes me think that the IQ250 has some serious color stability problem in shifted areas, due to it's strange angular response.

For the IQ250 to be a real option for someone with similar shooting style to mine, it must handle ~32mm with 10mm shift, or else I'm it does not go wide enough. I have not been able to see if that is possible or not with maintained color fidelity. I also think that Rodenstock 32mm on a 44x33 sensor is overkill (especially if only a limited part of the image circle is usable), but if you're into a back which is $35K I guess you could shell out $7K for a 32mm lens too.

This sensor has extremely low read noise, meaning that even if we lose 4 stops in the shifted areas we don't get much noise. However I'm a bit concerned that with so few photons captured that we could lose color fidelity for that reason too. Pushed dark areas are rarely colorful even if they are low noise.

 

[Paul2660]

The results from the 32mm Rod show that a CF would benefit both cameras. I agree that a blue sky shift would be a great test as it's hard enough to balance on the 260 at times when shifting with no clouds.

I need to look at the 60mm and 40mm Rod shots.

Torger, did you look at the 40mm so see if has the same color issues you see on the 32mm?

Paul C

 

[torger]

Torger, did you look at the 40mm so see if has the same color issues you see on the 32mm?

I looked at it now, and if it's there it surely much smaller. So I don't think I see any color issue on the 40mm, but as said this type of ceiling is more forgiving than an overcast sky.

What I did see on the 32mm JPEG is demosaicer failure, look just above the "15 mm Rise" label were you see a stitch seam, the topmost image has quite severe mazing and desaturated colors. I think this is due to the different angular responses of the two green channels, and the LCC has not succeeded balancing that out. Possibly with improved LCC algorithm the demosaicing failures and color issues could go away.

A full-frame CMOS with identical sensor technology like the IQ250 would probably work with the 40mm, and fullframe+40mm is wide enough for many. Let's hope such a back appears soon. If we're lucky they make pixel size a little larger so we get better angular response too. 60 megapixel full-frame CMOS would be nice.

 

[AreBee]

Torger,

...look just above the "15 mm Rise" label were you see a stitch seam...

Why would a seam be visible, given that the back alone moved between captures?

 

[Stefan Steib]

Thanks a lot Doug for that good piece of work.

Very interesting ! One thing that is definitely not expected (well I suspected it but now there is proof) - the CMOS beats the CCD´s in about any aspect besides that the IQ280 is bigger.
The IQ260 is a disappointment, grainy, bad shadow details, color noise and actually less quality than the 50 Mpix at 100 ASA though it was shot at 50 ASA. The IQ280 is on 35 ASA and just reaching the level, probably better with more light and less critical brown tones (compliment Doug for this honesty. It is the ultimate test - brown with lowlight shows everything !)

One hint: the standard settings in C.O. 7.2 are oversharpened (for all 3) and the 260 needed addition of moire.

This needs to settle down a bit, but in my brain it already forms a BIG question: if the CMOS is SO much better, who will stil buy a CCD ?

? ? ?

Greetings from Germany
Stefan

 

[Paul2660]

Torger,

You have a good eye. I still can't see the stitch line, but on reading you post on the 250 and color sat, I missed that as I didn't look at the finished image or the top images. You are exactly right, the color saturation does fall off, and it's especially noticed on the brown elongated ovals towards the top center.

What's also is that the stitched 260 image has a strange noise blob on the large partition. The top 1/3 of the partition is actually OK to me, but the lower 2/3's, areas I saw in looking at only the bottom row of images, is way to full of noise. Here is quick shot.

 

[torger]

Why would a seam be visible, given that the back alone moved between captures?

Even if alignment is perfect a seam can be visible if the two stitched images has different noise levels, ie you see a line where the noise suddenly increases. In this case there is a misalignment though. In the 32HR IQ250 stitched jpeg just below the "15mm Rise" label you see a alignment error (below the "15" before the 60mm IC line, ie just 35 pixels below). I stated incorrectly in my previous post that it was directly above.

So why is there an alignment error? The camera might have been ever so slightly moved, turning the knobs on the camera body to shift the back can be hard without disturbing the the setup the tiny amount needed to cause a few pixels misalignment.

A state of the art stitching algorithm can stretch images to make them fit anyway, but I would guess that there is some simpler stitching algorithm used here that don't stretch the images -- and surely we would not want that in this case as we want to pixel peep and see the true pixels the back has captured and not some possibly slightly scaled stretched and rotated image.

 

[torger]

Torger,

You have a good eye. I still can't see the stitch line, but on reading you post on the 250 and color sat, I missed that as I didn't look at the finished image or the top images. You are exactly right, the color saturation does fall off, and it's especially noticed on the brown elongated ovals towards the top center.

What's also is that the stitched 260 image has a strange noise blob on the large partition. The top 1/3 of the partition is actually OK to me, but the lower 2/3's, areas I saw in looking at only the bottom row of images, is way to full of noise. Here is quick shot.

I updated in a post above about where the seam is, my first post was misleading I said above instead of below, sorry!

Concerning the noise blob I think you see a seam between two images. The images overlap a lot so the stitching algorithm don't have to make a straight seam and it has not done it here, so the noisy part is from a noiser image which has been stitched in.

The question is why so large difference in noise when exposure time has been the same for all images? My guess is that there's some heat problem involved here and the IQ260 is performing much worse than it should, and obviously varying in performance between images. The high noise levels and large amount of hot pixels seen is so surprising I think something is wrong with the back or how it was used in the shoot.

 

[torger]

Another issue;

To make LCC correction the algorithm must have a reference color, ie it must have some point in the picture which is neutral. However with extreme shifts, like in the 30mm upwards shift, there probably is no part of the picture that is neutral for the IQ250, ie the whole picture has color cast. This means that the LCC algorithm cannot know what neutral is and will probably fail to develop the same color balance as the unshifted image.

The desaturation of the ceiling we see can be due to this.

The demosaicing failure (mazing) seen in parts of the ceiling is due to that the LCC algorithm fails to balance out the G1/G2 differences that occur due to color cast, this could also be a side effect from that the whole image has color cast.

I think it could be technically possible to get better IQ250 results than Capture One can do now, but you need a better LCC algorithm.

 

[Paul2660]

I am used to seeing the loss of Saturation in extreme shifts, as I believe it's just do to the limit of the lens in question image circle.

For example the SK35, when shifted on the IQ160 or any 60MP begins to show this at around 9mm of shift and by 12mm the image is basically close to B&W, not to mention the extreme magenta color.

The corresponding LCC will correct actually quite a bit of the magenta shift, surprisingly so, but it's can't bring back the loss of saturation.

What you caught in the top of the ceiling shot to me is just the IQ250/32mm Rod limit of rise due to the image circle. The IQ260 does better here, but it's a larger sensor which may account of the differences.

What I see in the ceiling shot with the IQ250, is no extra detail smearing or noise, (which you can get with the IQ260) just loss of saturation and that can be recovered.

Your point of a pure blue sky however is very good one, and I don't know how well it would recover as here you are trying to blend back on pure blue and that to me is one of the hardest things to do. If there were clouds in the sky to break up the blue, it would be much easier.

It would also be interesting to see if the CF makes any difference on this as it would most definitely make the LCC have to work less to recover.

Paul C.