PhaseOne Sensor Plus Observations and Questions

[Ken R]

Hi, I own a PhaseOne IQ160 and one of the features of the phase back is the sensor plus setting. At first I though I was not going to use it and for the main application for which I got the back for I have not. But recently I have experimented using the back (mounted on the H1 instead of the Arca) just like I would my DSLRs, handheld and in low light conditions for photographing people. I have to say I was impressed. I used sensor plus mode at iso 400-800 and the results were really a revelation.

The first thing I noticed when opening the files on my mac was that they looked like they had more resolution than the 15MP size suggests. Any explanation to this? I mean 15MP is 15MP, I know per pixel quality varies from system to system but this was well above what I expected. The detail is just there, more natural, no jaggies that I could see. The files are also very clean, what noise is there looks really good, smooth, natural, hard to explain.

The other thing I noticed happened because I accidentally overexposed a few images when the light quickly changed but I kept shooting a few frames before I adjusted the settings of the camera. I opened the images on C1p7 and expected them to be trash with blown highlights, yes, the subject had a white shirt!, but no, I was able to recover the highlight detail on the shirt. This really, really surprised me. I have worked with a LOT of different DSLRs and never seen a highlight recovery this large. Again, this happened while using sensor plus. Any explanation is appreciated.

Also, I never had to wait for the back, I just shot frame after frame. The AF on the somewhat clunky medium format SLRs is not what you call quick anyway and shutter lag and blackout is more than what you get with even a basic dslr. AF performance with the H1 was good though even in low light. I got a pretty high percentage of well focused shots.

[Paul2660]

Ken:

I have noticed at 400 bit more room in highlights as you mentioned with sensor plus.  The 15mp is sharp also and at first I thought it would uprez to 30MP or so pretty easily, however I did quite a bit of work on files over the summer and really found that once you took them up much past 25% then the usual issues showed back up.  I have created several nice multi part panos with the 15MP files and they look very nice.  As you pointed out the files are very clean and very crisp.  You can also pull up the shadows nicely at iso 400 in sensor plus, much better than regular iso 400.

The 20Mp output from the 180 is one thing I would love to have, but I will be happy with the 15MP.  It's definitely a nice feature to have in the kit. 

Paul Caldwell

[eronald]

Every digital camera has a point at which ISO is "fake". For digital backs this point is ... always.

So it's impossible to overexpose a DB by something 2 stops too bright when shooting ISO 400, because the back is actually always set to 50 or 100.

For a dSLR this situation occurs when the amps are maxed out.

Edmund

[Paul2660]

Edmund

You bring up one of the real grey areas of the Medium format backs.  I understand it the same as what you stated as I believe it's true for any CCD chip.  Where as you can push the gain on a CMOS chip with the iso setting.  The Phase One backs are shooting pretty much at their base even when you push them up. Best case is base plus one stop so your real iso range would be iso 50 and 100.  Thus, as I understand it, the results of a 2 stop underexposed iso 50 shot should be the same as a properly exposed iso 200.  For while on LuLa Wayne Fox had a post that showed a good example of this I believe from a IQ180.  Wayne I hope I have the order right, I remembered your post and the images you had which were a very good presentation. 

However in my real world shooting, with Phase One backs, I still tend to find the results at least up to iso 200 are best if you pick the iso that works for the required exposure instead of underexposing iso 50 in this case.  The shadows will have more noise in the underexposed iso 50 image.  This is not the same as underexposing  Nikon CMOS D800 at iso 100 and then pulling up the shadows as they will be excellent as you can easily pull up a D800 image 2.5 stops at iso 100 with almost no noise in the shadows.   In my use at iso 50 you can get about 1.5 to 1.75 of a stop before the shadows start to show excessive noise.

As Ken pointed out, the sensor plus image can be a very good file and often can save the day when the conditions go bad.  I can't remember if sensor plus is allowed at iso 200, but I know it goes from 400 on up.  My net rule is in low light or low light plus wind where I need a faster shutter speed (normally I would be looking for a higher iso) I will switch to iso 400 with sensor plus and forgo the 60MP res.  The files are very good and can be uprezed  up in size a good bit since they are so clean.  Where as the same 60mp iso 400 image can tend to be noisy, with a good amount of color/sat fall out.  Not true with the sensor plus file.  You may have have to down rez the 60mp iso 400 image a bit due to the noise.

The part of the equation that I still trying to understand is just what is happening when the IQ260 goes into Long Exposure Mode, and the base iso is now iso 140.  This is a pretty good push from 50 and so far I have had very good results at iso 140 and 200 in Long exposure mode at full resolution on the 260. 

Paul Caldwell

[Ken R]

Every digital camera has a point at which ISO is "fake". For digital backs this point is ... always.

So it's impossible to overexpose a DB by something 2 stops too bright when shooting ISO 400, because the back is actually always set to 50 or 100.

For a dSLR this situation occurs when the amps are maxed out.

Edmund

I know, but there is something else going on because the image definitely looks much better than if I used the base iso (50) and just raised the "exposure" in software later. (which I have done in testing on previous occasions)

[torger]

Large highlight recovery simply means that the sensor is more underexposed, ie the back rendering and capture one chooses to render the image brighter than the typical DSLR would. There's no magic going on.

If you use a camera-agnostic raw converter like RawTherapee which handles all files the same regardless of what camera model it is you can see the true exposure of the file, ie these files will look a bit dark before pushing it.

A problem with the most popular raw converters like capture one and lightroom is they do automatic stuff even with neutral settings. The intention is to simulate film behavior and make the software easier to use, which it indeed becomes, but the disadvantage is that if you want to really know what the exposure is, test highlight recovery and do comparisions between camera models etc you can't trust the results as the hidden automatic adjustments may differ depending on camera mode, model, exposure etc.

If you want to do the test if ISO50 pushed 2 stops is equal to ISO200 I think you'll have to use RawTherapee or other raw converter which can render images more neutrally than Capture One or Lightroom can. Unfortunately I think the sensor plus mode is not well-supported there, but I don't know for sure. Looking at DxOmark data it seems like it could be the case, as one stop ISO increase shows one stop DR decrease, on the other hand many DSLRs show this behavior too, including the D800 so it's not a safe way to check if scaling is analogue or digital.

I'm sure though that even if all processing of sensor plus is digital the result will be better than scaling up ISO50 and downscaling resolution, as you can do better use of the sensor idata f you know it will be used in half size. I'd guess Phase One has put in quite some effort to make it look best possible, and this is a mode which is not available in other cameras or raw converters afaik. Canon has something similar with their mRAW as sRAW modes, but it's implemented much differently, their purpose is not to extend ISO but just to reduce raw size when you don't need the full resolution.

[yaya]

Every digital camera has a point at which ISO is "fake". For digital backs this point is ... always.

Even DxOMark think otherwise....http://www.dxomark.com/Reviews/DxOMark-review-for-the-Leaf-Aptus-75s

[Theodoros]

Is the sensor+ a "true colour" mode? I mean is the integrated outcome of the four pixels used a pre-processed algorithm of 2xgreens +1red+1blue that leads on each of the "new" 15mp pixels to carry "true colour" (other than green or blue or red) wavelength?

[eronald]

Even DxOMark think otherwise....http://www.dxomark.com/Reviews/DxOMark-review-for-the-Leaf-Aptus-75s

Yair -

 I eat my words The Leaf 75s has variable gain amps.
 Are there any Phase backs with those too?

Edmund.

[Steve Hendrix]

Large highlight recovery simply means that the sensor is more underexposed, ie the back rendering and capture one chooses to render the image brighter than the typical DSLR would. There's no magic going on.

If you use a camera-agnostic raw converter like RawTherapee which handles all files the same regardless of what camera model it is you can see the true exposure of the file, ie these files will look a bit dark before pushing it.

A problem with the most popular raw converters like capture one and lightroom is they do automatic stuff even with neutral settings. The intention is to simulate film behavior and make the software easier to use, which it indeed becomes, but the disadvantage is that if you want to really know what the exposure is, test highlight recovery and do comparisions between camera models etc you can't trust the results as the hidden automatic adjustments may differ depending on camera mode, model, exposure etc.

If you want to do the test if ISO50 pushed 2 stops is equal to ISO200 I think you'll have to use RawTherapee or other raw converter which can render images more neutrally than Capture One or Lightroom can. Unfortunately I think the sensor plus mode is not well-supported there, but I don't know for sure. Looking at DxOmark data it seems like it could be the case, as one stop ISO increase shows one stop DR decrease, on the other hand many DSLRs show this behavior too, including the D800 so it's not a safe way to check if scaling is analogue or digital.

I'm sure though that even if all processing of sensor plus is digital the result will be better than scaling up ISO50 and downscaling resolution, as you can do better use of the sensor idata f you know it will be used in half size. I'd guess Phase One has put in quite some effort to make it look best possible, and this is a mode which is not available in other cameras or raw converters afaik. Canon has something similar with their mRAW as sRAW modes, but it's implemented much differently, their purpose is not to extend ISO but just to reduce raw size when you don't need the full resolution.

One note about Capture One and automatic adjustments as files are imported. You can view a pre-import histogram via the Exposure Evaluation tool. If you compare the Histogram tool with the Exposure Evaluation tool after importing - even with no adjustments made at all - there is usually a difference in the 2 rendered histograms.


Steve Hendrix
Capturer Integration

[Theodoros]

Still wonder if the exposure+ mode is only a sensitivity benefit or if it's additionally a "true colour" mode…

[EricWHiss]

Doesn't the Sensor Plus use some kind of pixel binning on the chip so they can read 4 pixels once thereby quartering the read noise?

[torger]

Yes there's some sort of binning going on, and this binning combined with optimized AD conversion makes it a bit better than the result would be if just downscaling a full resolution image. I have not been able to find a detailed technical description of how it works though, it's outlined here in DxOmark's look at P65+ http://www.dxomark.com/Reviews/DxOMark-review-for-the-Phase-One-P65 but their description of it is based on what they see in measurements, from Phase One themselves I've only seen marketing type of descriptions.

Concerning color, with a 1/4 resolution you get two green, one blue and one red pixel binned into one. However they cover different parts of the pixel so there's still not true color compared to a multi-shot back where you get full coverage of the separate color filters in each pixel. When pixel peeping you still might see less color artifacts of details than you would a full resolution image though.

[Theodoros]

Concerning color, with a 1/4 resolution you get two green, one blue and one red pixel binned into one. However they cover different parts of the pixel so there's still not true color compared to a multi-shot back where you get full coverage of the separate color filters in each pixel. When pixel peeping you still might see less color artifacts of details than you would a full resolution image though.

There could be an average of the four pixel info that would lead on kind of "true colour"… Surely the output is not as good as my 528c shot MS, but when I rented a P65+ back on 2011, I did notice a slight colour improvement when it was used in exposure+ mode, if it is not a "true-colour" output though, how do they determine the (new) G-R-G-B Bayer pattern output sequence, since all "new" pixels have 2xG + 1xG + 1xR… it's a "head scratch" situation, ...isn't it? May be Doug could help here, as you say, official P1 information comments on noise improvement and sensitivity increase only.

[Doug Peterson]

Sensor+ does NOT use a traditional binning arrangement. It's a patented sampling pattern hardwired into the sensor design and is an excellent way to downsample from a larger resolution sensor. I use sensor+ for maybe 75% of the images I shoot at a wedding.



Phase One Sensor+ PDF Explanation

[torger]

Thanks for the doc, most appreciated by us tech geeks!

The most obvious way to us that make software and not hardware would be to take 2x2 pixels and just make a full-color RGB super-pixel from that, I assumed it was something like that in my earlier posts in this thread. That would work fine, but the problem with that which I realize now when reading the document above is that you would then need to read each individual pixel from the sensor, ie no binning, ie no performance increase in terms of ISO, you could then just make a full-resolution image instead.

And when you put four pixels into one A/D conversion to reduce read noise, you can't read mixed colors, ie you must sample 4 pixels of the same color. And then you get real problems with resolution, as shown in the document. Phase One gets around that by a special sample pattern. You don't get a new full-color super-pixel from 4 small pixels, but instead 4 new super-pixels (1 red, 1 blue, 2 green, as usual) from 16 small pixels, and do demosaicing on that.

The attached image from page 3 is the key, which shows how 16 small pixels is converted to 4 super pixels, which is made in 4 instead of 16 A/D conversions, reducing read noise. Probably they've made a custom demosaicer to handle the ortogonal arrangement of pixels, I would guess a standard demosaicer would produce artifacts as they don't expect the ortogonal arrangement of green.

It's great to see this document as it shows that there's really a hardware achievement and that sensor+ is not just some simple hack in software, which one can suspect when just reading the marketing descriptions.

[Theodoros]

+1

[ondebanks]

The part of the equation that I still trying to understand is just what is happening when the IQ260 goes into Long Exposure Mode, and the base iso is now iso 140.  This is a pretty good push from 50 and so far I have had very good results at iso 140 and 200 in Long exposure mode at full resolution on the 260. 

Paul,

When you raise ISO on any camera, you record less light [faster shutter speed and/or smaller aperture, for a given metered eV] and use less of each pixel's capacity.
So ISO 140 is used as the minimum ISO in IQ260 LE mode because the sensor must reserve a large chunk of its pixel well depth for thermal dark noise isolation. Dalsa/P1 cannot let you fill up the pixels with photoelectrons as you would when shooting the same metered eV at ISO 50.

As with any ISO above base on a DB, you are technically underexposing the image at ISO 140 also. They may apply digital gain prior to the RAW writing, or else in C1, to make it seem correctly exposed. [OTOH, in the digital realm, what is "correct" exposure anyway? Once you're operating at high signal to noise, that's "correct" enough].

If you are comparing your IQ260 at ISO 140 to your old P45+ at ISO 50 and thinking they're about equal on noise, despite the ISO difference, that is down to two main quantitative factors (plus less quantitative ones that Doug tends to point out, like newer/better ASICs and firmware). Firstly, the IQ260 is inherently more light sensitive in a give exposure time due to higher quantum efficiency - Kodak's "naked" CCD tech, as used in the P45+, is more efficient that Dalsa's, as used in say the Aptus 33; Kodak's microlensed CCD tech is also more efficient than Dalsa's microlensed CCDs; but in this case you are comparing a microlensed Dalsa IQ260 with a non-microlensed Kodak P45+, and so the Dalsa has the efficiency edge. Secondly, the IQ260's CCD has lower readout noise - roughly 11 electrons vs 16 electrons, if they adhere to the datasheets regarding readout speed. More signal and less noise brings ISO 140 close in appearance to the ISO 50 of the other back.

Ray

[Paul2660]

Ray

Thanks for that detailed response.  That does help.  I have always assumed the 60 and 80 MP backs from Phase One did not have micro lenses like the older Kodak chip in the P30+.

Paul

[ondebanks]

Even DxOMark think otherwise....http://www.dxomark.com/Reviews/DxOMark-review-for-the-Leaf-Aptus-75s

That link says: "Our measurements show that the ISO Sensitivity settings of the camera are real (that is, all ISO Sensitivity settings are obtained by applying a gain, analog or digital, before RAW delivery) for the entire ISO Latitude range of 50 to 800."

Doesn't that mean that the Aptus 75s needlessly throws away up to 4 stops of overexposure headroom (DR) above base ISO? Why would Leaf have taken that decision?

Compare to DXoMark on the Hasselblad H3Dii-50 [and H3Dii-39]: "With respect to ISO sensitivity, the only real ISO provided is ISO 50 (measured sensitivity is 45). All other ISO values are obtained by applying a digital gain during the RAW-to-JPEG conversion, which is the same type of design as for the H3DII 39."

DxO then make the mistake of extrapolating (their terminology) rather than measuring the Hasselblads' DR at higher-than-base ISOs, and they presume that it falls by 1 stop per ISO stop. This is incorrect. SNR will fall with ISO, just as they extrapolate, but there is no reason why DR will. The readout noise has not changed. The signal capacity of each pixel has not changed. For each stop of shadow detail lost, a stop of highlight detail is gained, moving below the saturation threshold. It doesn't matter whether the scene being photographed contains subject matter of that brightness or not. This is about sensor DR, not scene DR.

As for Phase One P45+: "ISO latitude ranges from ISO 50 to ISO 800, but measurements show that the only two first ISO values (50 and 100) are real, a digital gain is applied during RAW conversion from ISO 200 to ISO 800".
So it's interesting that for the most part P1 take the Hasselblad approach, except at ISO 100.

Again, "Curves and results for “non-real” ISO are extrapolated and shown with dotted lines on dxomark.com" for the P45+...so again their DR curve is wrong, above ISO 100.

The later PhaseOne backs (P65+, IQ180) take a sort of hybrid approach: at moderate ISOs up to ISO 200, you retain the DR and keep gaining the headroom, like the Hasselblads; but then at higher ISOs they lock that headroom gain in and let the DR shrink - "The Phase One P65’s ISO latitude in RAW shows a design change when compared to the previous Phase One model:, the latest P45+ provides only two real ISOs, ISO 50 and ISO 100; by contrast, the P65+ provides real ISO from 100 to 800 (in high resolution mode) and from 400 to 3200 (in Sensor Plus mode). ISO 50 (high resolution mode and ISO 200 (Sensor Plus mode) are obtained only by applying a digital gain during RAW conversion."

Ray