Can we expect new sensors at Photokina?

[ErikKaffehr]

Hi Steve,

Excellent summary, seen from a technical/economical standpoint.

Best regards
Erik

I believe there will never be a backing up of resolution for medium/large format.

I do hope they keep photosite size where it is or at least don't go too much smaller. I wouldn't rule out that it may go bigger, but if so, I would expect by very little. We have customers using digital backs with 9 and 12 micron photosites who can't shoot higher than 50 ISO without gaining unacceptable levels of noise, color artifacting, etc. Of course that is older technology, but other than Canon with the G11/S90, I can't think of anyone who has backed up on resolution.
The resolution issue is so interesting because I hear so often how "who needs so and so many megapixels, stop already". But I've been hearing this ever since the 22MP DB's came out. And yet, the P65+ has been a huge success.

Canon and Nikon will continue to up the megapixel count, and yet when it is discussed as in rumor about the next model, it's usually as a positive "The 1DS-MKIV is supposed to be 32MP's!". There's very little negativity associated with it....yet.

And for sure that is because Canon/Nikon do so much more with their cameras and in ways that more effectively benefit the demands of commercial photography today. So medium format is seen as concentrating too much on megapixels, size of file, etc, rather than making their cameras more flexible and versatile, ala 35mm. But currently, that is one of medium format's few remaining advantages over 35mm. I can guarantee you that a $28,000 6x7, 24MP digital back will fail in the market place. While some may see advantages in the lower resolution, the majority will pass at paying that premium for a product that captures only 2/3 the file size of 35mm.

I agree and believe that medium format has focused on resolution and chip size enhancement (though modestly), but more because other enhancements take a long time. I know they have been working on them. But in the meantime, you can't fault a company for putting out products that people buy and that do produce revenue. They do have to make money while they work on technology that may be 4-6-8 years away. It kind of puts a drain on R&D if you don't have revenue coming in.

Compared to 35mm, medium format has always had the advantage of bigger imaging area, larger image/file size. Those have been their primary advantages. And those advantages were traditionally necessary and a clear reason to choose them. With the changes in the commercial markets, the type of photography being demanded in general (even with wedding, portrait, architecture, etc), the need for speed, flexiblility, maneuverability, and with 35mm now producing (somewhere) in the range of traditional medium format image quality, (and not to mention crimped stills budgets) medium format's big advantages have lost their influence with commercial photographers (though upping it with other markets that value those assets and appreciate the evolution medium format has made from film-based to digital).

So...I expect that we will see more resolution announced this fall, hopefully spread over a larger sensor size, and hopefully progress will have been made on the issues medium format struggles with - usable LCD, in-camera functionality, etc. I see this as a benefit to those in the commercial field (and elsewhere) who still choose to use medium format  for its strengths. It helps make that choice to remain with medium format easier, which is a very big plus. For those who don't benefit from medium format's strengths, it probably doesn't matter how improved in terms of "usability" medium format gets because it is, at least for the time being, not the right product for them and that choice has already been made.

We'll see.


Steve Hendrix

[rainer_v]

i agree steve, very good written and analysed. and very honest.

[BJL]

One more thumbs up for Steve's discussion. It seems that the MF backs with higher pixel counts keep being the most profitable: Steve implies that they sell well by MF standards, and the price premium for more MP seems to exceed the extra component costs, so the markups are probably highest for the higher MP models. Given that message from the marketplace, my guess even more is that the main change in MF sensors this year is likely to be larger sensors at the top of the line, with the same 6 micron pixel size ... and this is the best that those who fear "excessive pixel counts" can hope for.

[eronald]

I like Steve's honesty -but have a minor remark:  Nikon has had good success pitching the large-pixel D700 against the 5DII which outresolves it by a factor of 2 (2 generations of sensors).

Edmund

One more thumbs up for Steve's discussion. It seems that the MF backs with higher pixel counts keep being the most profitable: Steve implies that they sell well by MF standards, and the price premium for more MP seems to exceed the extra component costs, so the markups are probably highest for the higher MP models. Given that message from the marketplace, my guess even more is that the main change in MF sensors this year is likely to be larger sensors at the top of the line, with the same 6 micron pixel size ... and this is the best that those who fear "excessive pixel counts" can hope for.

[EricWHiss]

Hum... companies design for the benchmarks their products will be submited to. The only DR benchmark I know of is DxO and their results are discarded as soon as they don't match people expectations...

In the end DR claims will end up being perceived as marketing claims since we have de facto collectively decided to give up on DR measurments.

The logical consequence of our choices will be a reduced focus from manufacturers on DR.

Cheers,
Bernard

 DXO isn't doing a great job with respect to measuring DR for MF DB.  As they sell their own software, it wouldn't be to their advantage to measure how much better a MFDB file gets when run through its own manufacturers software.  That's the part they have left out and its an important part of the MFDB process.    I don't take any stock in DXO measurements or those on other sites where they measure DR with the IEEE specs and not ones useful to photographers.  

I still would like to see more DR from the next backs but wonder if its possible. There's hardly been any change in DR of the digital backs (at base ISO) for a number of generations.  

[JSK]

Let's say the new size is it's P90+ with 120MP    how far are we compared to 8x10 Format?

any info.. Dalsa or Kodak?

[JoeKitchen]

Just looked at the Kodak webpage, their largest sensor is 50 mp with a 6 micron pixel, same dimensions as the P45 sensor.  Maybe we will see a P50? as a true upgrade from the P45?

[ejmartin]

DXO isn't doing a great job with respect to measuring DR for MF DB.  As they sell their own software, it wouldn't be to their advantage to measure how much better a MFDB file gets when run through its own manufacturers software.  That's the part they have left out and its an important part of the MFDB process.    I don't take any stock in DXO measurements or those on other sites where they measure DR with the IEEE specs and not ones useful to photographers.  

I still would like to see more DR from the next backs but wonder if its possible. There's hardly been any change in DR of the digital backs (at base ISO) for a number of generations.

DxO provides the information you seek, if you know where to look.

http://luminous-landscape.com/forum/index....c=42158&hl=

For image quality to improve in the most useful part of the range, sensors need to have higher quantum efficiency (the percentage of incident photons that are recorded by the sensor) and lower read noise.  It is on the latter score that MFDB's are woeful relative to contemporary CMOS-based DSLR's.

[JSK]

Just looked at the Kodak webpage, their largest sensor is 50 mp with a 6 micron pixel, same dimensions as the P45 sensor.  Maybe we will see a P50? as a true upgrade from the P45?

I like the idea of having good quality long exposures even though i don't use it daily.. I was looking at P65+ Dalsa backs and heard they are limited to 30 seconds or so..

Then I read about stacking Long Exposure images to make it look like 5, 10, 15, 30 minutes exposure.. but there is limited or no info/tutorial how to do it successfully..

[EricWHiss]

For image quality to improve in the most useful part of the range, sensors need to have higher quantum efficiency (the percentage of incident photons that are recorded by the sensor) and lower read noise.  It is on the latter score that MFDB's are woeful relative to contemporary CMOS-based DSLR's.

Well MFDB can't be all that woeful because they still have stops more DR than CMOS DSLR's in that kind of test.  What's that chart show only 7.5 stops for the nikon D3x at base iso and about 4 at 6400 iso?  Test any MFDB using the same S/N acceptance ratio and you'll have more like 10.5 or 11 at base ISO.

It looks like an increase in well size is going to help with read noise and that's what you see if you look through the sensor spec sheets for sensors designed for scientific uses (like astronomy) that have big wells.  Some of these have a SNR spec of 90 db.    But really, I was wondering if it would be possible to get much more DR without going to higher bit files and chip architecture.  It seems possible but in practice it hasn't been happening.

[ejmartin]

Well MFDB can't be all that woeful because they still have stops more DR than CMOS DSLR's in that kind of test.  What's that chart show only 7.5 stops for the nikon D3x at base iso and about 4 at 6400 iso?  Test any MFDB using the same S/N acceptance ratio and you'll have more like 10.5 or 11 at base ISO.

Sorry, wishful thinking.  If one uses the same criterion (range of EV over which the SNR is greater than 24dB) as in the example in the post I linked to above, the P65+ only gets about 7.0 stops DR at base ISO.  Again this can be read off the relevant SNR plot at DxO's test site

Now, that is DR per pixel.  But DR is depends on scale in the image, since SNR is typically higher on coarse scales and lower on fine scales (eg, the more you magnify an image, the more apparent noise becomes, and thus the lower the S/N ratio).  To fairly compare, one should look at DR at a fixed spatial scale, which means scaling the result by the square root of the pixel counts.  sqrt[65/24]~1.65 or about 0.7 stops extra for the P65.  In other words, the ISO 100 DR of the P65+ at the spatial scale of D3x pixels is about 0.3 stops better.  A third of a stop, that's all.  Not so great, consider it is gathering twice as much light to begin with (all other things being equal that would mean it should have a half stop advantage rather than a third stop).

It looks like an increase in well size is going to help with read noise and that's what you see if you look through the sensor spec sheets for sensors designed for scientific uses (like astronomy) that have big wells.  Some of these have a SNR spec of 90 db.    But really, I was wondering if it would be possible to get much more DR without going to higher bit files and chip architecture.  It seems possible but in practice it hasn't been happening.

Read noise depends on all sorts of design choices, and is not necessarily correlated to well capacity.  I could give examples if you like.  I'm not sure it's germaine to bring in astro CCD specs, that's a quite different set of design criteria for image capture.  As far as bit depth is concerned, one only needs as many bits as there are stops of engineering DR.  Present-day MFDB's need no more than 12 bits per pixel, perhaps a tad more if oversampling confers an advantage (though I have yet to see such).

[ErikKaffehr]

Hi,

The DxOmark data is based on raw data without conversion. If you look at the data Emil referenced you will find that the P65+ has a bit less DR at 24 DB SNR than the D3X.

Also, just realize that would MFDBs have a real advantage in DR it would also mean that they would perform well at high ISO. It seems obvious that they don't do that.

Emils explanation of pattern noise is interesting. Also note that in DxO tests Phase One P65 is a little bit better than Nikon D3X and Nikon D3X is much better than Canon 1DsIII,  Canon 5DII or Sony Alpha 900. Also note that the D3X probably achieves it's rating in the slow 14 bit/pixel mode.

Check also this: http://www.dxomark.com/index.php/en/Our-pu...e-Phase-One-P65.

That said, many renown photographers claim significant advantage of MFDBs over DSLRs.

Astronomy is more about collecting light than resolution. They normally photograph stars which are single point light sources. I guess that the sensors are cooled, probably with liquid nitrogen.

Best regards
Erik

Well MFDB can't be all that woeful because they still have stops more DR than CMOS DSLR's in that kind of test.  What's that chart show only 7.5 stops for the nikon D3x at base iso and about 4 at 6400 iso?  Test any MFDB using the same S/N acceptance ratio and you'll have more like 10.5 or 11 at base ISO.

It looks like an increase in well size is going to help with read noise and that's what you see if you look through the sensor spec sheets for sensors designed for scientific uses (like astronomy) that have big wells.  Some of these have a SNR spec of 90 db.    But really, I was wondering if it would be possible to get much more DR without going to higher bit files and chip architecture.  It seems possible but in practice it hasn't been happening.

[ErikKaffehr]

Hi,

Two reasons probably.

1) Nikon doesn't have a 24 MP model below USD 3000

2) Utilizing those 24 MPixels take a tripod and good technique

Best regards
Erik

I like Steve's honesty -but have a minor remark:  Nikon has had good success pitching the large-pixel D700 against the 5DII which outresolves it by a factor of 2 (2 generations of sensors).

Edmund

[eronald]

I have another thought for you: if Nikon priced its 24MP product the same as the 12MP - which they almost could, same body, same chip size and yield, same memory bandwidth- their press  customers get confused, buy it, and suddenly get noisy unfocused images at pixel level and the product wouldn't sell.  This way they customers get a camera which is very fast in frame rates, focuses perfectly, and has ISO of 6400 no sweat - then the customers don't *try* to buy better.It's an astute marketing trick, steering customers to the camera they *need* by pricing it lower.

When I image a fashion show with the D3x, I know the guys with the D3 will get better results.

Edmund

Hi,

Two reasons probably.

1) Nikon doesn't have a 24 MP model below USD 3000

2) Utilizing those 24 MPixels take a tripod and good technique

Best regards
Erik

[BernardLanguillier]

That said, many renown photographers claim significant advantage of MFDBs over DSLRs.

I hate to have to write this because I don't really care, but:

- Few, if any, use the D3x,
- These claims are never backed up by any data or sample images for that matter.

What many backs users are seeing are backs that are calibrated for under-exposure, which gives them the impression that they have more highlight headroom for a given "correct exposure" (as defined by a traditional film based approach). More apparent highlight headroom is materialized by the mythical ability to recover blown highlights. I write "mythical" because it doesn't exist with the linear sensors we have today.

Highlight recovery can only be made possible by the existence of data in the first place, which is only possbile because the back histogram doesn't show the real information available in the files. Put it otherwise, a digital camera equiped with a linear sensor (read 100% of sensors on the market today) can only show highlight recovery ability if the system is calibrated for under exposure.

In other words, any DR claim based on apparent highlight recovery ability alone is misguided in the first place.

Any DR comparison has to be made based on actual raw ETTR (regardless of exposure and of in camera histograms) and can only be made by looking at how noisy the shadows are.

Now this comparision can be done and might show some advantage to the backs, but we are getting awfully close to the DxO results and the only possible difference between their results and actual prints comparision is in the "look of the noise".

If this has been done and showed results differing significantly from DxO results, how come we don't see these results? Again, don't show me 5DII results, I know about the noise banding at ISO 100 in deep shadows.

More often than not things that are hard to find end up not existing at all...

Cheers,
Bernard

[EricWHiss]

Sorry, wishful thinking.  If one uses the same criterion (range of EV over which the SNR is greater than 24dB) as in the example in the post I linked to above, the P65+ only gets about 7.0 stops DR at base ISO.  Again this can be read off the relevant SNR plot at DxO's test site

Now, that is DR per pixel.  But DR is depends on scale in the image, since SNR is typically higher on coarse scales and lower on fine scales (eg, the more you magnify an image, the more apparent noise becomes, and thus the lower the S/N ratio).  To fairly compare, one should look at DR at a fixed spatial scale, which means scaling the result by the square root of the pixel counts.  sqrt[65/24]~1.65 or about 0.7 stops extra for the P65.  In other words, the ISO 100 DR of the P65+ at the spatial scale of D3x pixels is about 0.3 stops better.  A third of a stop, that's all.  Not so great, consider it is gathering twice as much light to begin with (all other things being equal that would mean it should have a half stop advantage rather than a third stop).



Read noise depends on all sorts of design choices, and is not necessarily correlated to well capacity.  I could give examples if you like.  I'm not sure it's germaine to bring in astro CCD specs, that's a quite different set of design criteria for image capture.  As far as bit depth is concerned, one only needs as many bits as there are stops of engineering DR.  Present-day MFDB's need no more than 12 bits per pixel, perhaps a tad more if oversampling confers an advantage (though I have yet to see such).

I'm suggesting that you check DXO's initial assumption before cranking out a lot of science based on their data.  DXO is not using C1 to convert phase files first and that improves them significantly.   DSLR do a fair bit of file processing in the camera before the RAW is written.  MFDB handle this differently and do some final file processing in software such as black frame subtraction.   You just can't look at these platforms as the same.  If DXO makes the leap that they are then linking to DXO as reference isn't going to help.  Don't forget they are not a scientific body, rather they are a for profit company and one that happens to make software only for DSLR and not for MFDB.

[EricWHiss]

Hi,

The DxOmark data is based on raw data without conversion. If you look at the data Emil referenced you will find that the P65+ has a bit less DR at 24 DB SNR than the D3X.


Best regards
Erik

This is exactly my point - were DXO to repeat the test with the black frame subtracted and what ever else MFDB software is doing then the DR figures would change.   In any case it would be really exciting to see a larger sensor at photokina.

[rainer_v]

This is exactly my point - were DXO to repeat the test with the black frame subtracted and what ever else MFDB software is doing then the DR figures would change.   In any case it would be really exciting to see a larger sensor at photokina.

basically bernhard is 100% right which was he says. c1 is not changing the raw data, the teoretical dr is determined by the raw information and cant be increased by the sw, but it can be decreased by bad code.

highlight recovery is done by the software and can be done very different, this means the restoration of color- neutral data if one or two of the color channels still hold some  information, meanwhile one channel (or two) are clipped.
 this is not different if used mf data or 35mm data, how usefull it is depends on the sw and if it either clips the whole file if one channel is clipped or if it use the rest information and is able to get colorshift free highlights. if done bad this mostly results in cyan or magenta casts in these highlight zones.
it should not be valued by DXO, cause its not a real part of the DR, it depends too much on the kind of image if these infos can be restore and - as i wrote above - this has nothing to do with the sensor size.

[JdeV]

Let's say the new size is it's P90+ with 120MP    how far are we compared to 8x10 Format?

any info.. Dalsa or Kodak?

By what criteria do you want to make the comparison?

1) You get considerably more accurate colour from a P65+ than any form of scanned film, including 8x10.

2) You also get considerably less dynamic range (at the highlight end) than scanned colour negative (but significantly more than scanned colour transparency).

3) The P65+ will generate unpleasant moire under relevant conditions, it can also render neon and some other subjects in strange bad ways that are hard to correct.

In real world tests a P65+ with a good lens will generally yield similar or even slightly better levels of resolution than a scanned sheet of 4x5 film. However, 8x10 is still in another league. The increase in resolution going from 4x5 to 8x10 scanned film is only slightly less than a straightforward linear geometric increase. To equal 8x10 resolution, even going to a 5 micron sensor (about the limit with foreseeable lens technology), will require something of the order of 240MP and a sensor size of around 66mm x 90mm.

However, to put this in perspective, shooting 8x10 is a royal pain and very prone to technical errors which render its potential rather moot. Of course it also has very shallow depth of field so in most circumstances the resolution advantage only applies to shots focused at infinity without the requirement of sharp foreground information. Because of this and because of point 1) a digital back some way short of this size and resolution would still be effectively superior to even 8x10 film for virtually all purposes.

[Bart_van_der_Wolf]

This is exactly my point - were DXO to repeat the test with the black frame subtracted and what ever else MFDB software is doing then the DR figures would change.

Hi Eric,

While a blackframe subtraction would change the DR, it is not for the better. Simple blackframe subtraction is only for the removal of non-random (pattern) noise, it increase random noise levels (by up to 40% for the shadows).

The DxO database is not such a bad startingpoint.

Cheers,
Bart