DR and RAW bit depth on Canon's forthcoming...

[jrista]

With all the details flying around about Canon's forthcoming releases, I'm curious if there has been anything about dynamic range and RAW bit depth for anything announced or rumored? Despite the fact that Canon claims 14-bit capability, they have not seemed to make fully effective use of it so far, with most of their cameras topping out in the 11-12 stops of DR range. Does anyone know what the DR of the 1D X is...will it have a true 14 stops like most of the recent Nikon cameras? Have there been any rumors stating DR for the 5D II successor or this possible new 40mp model?

It would be fundamentally disappointing for Canon to miss the mark AGAIN on dynamic range and keep all their loyal customers bound to an 11.7-stop limit on DR when the competition is blowing that out of the water. Of all the various improvements Canon could make, having thought about it the last few days, I would take a 5D III with the same resolution as its predecessor, same ISO range, and only the 60D's 9-point AF...if I could have better dynamic range. An extra two stops would do more for my landscape photography than anything else...including unbelievably high ISO, the 7D or even 1D X AF, etc. I could stop wasting time setting up and adjusting the single 2-stop GND filter I usually have to use, and just spend my time capturing the beauty before me as-is.

[Curmudgeon]

jrista,
 I posted a very similar comment on the thread about reaction to the firmed-up specs of the 5D3.  (I too begged for two more stops of DR--fat chance.) To me the lack of dynamic range in the 5d2 is its most glaring deficiency-- and the 5D2 is probably one of Canon's better performers in this regard. And yet there seems to be very limited concern about dynamic range among the users of this forum.  I think I am the 393rd reader of your post, and the first to make a response.

For me DR is a constant problem--much more so than autofocus or high ISO performance. Any time there is a horizon in my composition it's a pretty good bet that something is either going to be blown out or blocked up. As you say, there are many occasions where neutral density filters, AEB/HDR, even a tripod, are not practical because of subject movement, location, etc. What's really disturbing is to compare my current efforts to some of my 40-year-old prints from 35 mm film. I don't expect two stops in the new 5D, but if there isn't at least one real stop, I'll spend my photographic budget on lenses. The 24L TS-E is looking mighty tasty.

I was going to open a thread on this subject myself, but I'm new to this forum and before flogging what I assumed was a thoroughly dead horse, I did a forum search. Until you started this thread I found incidental snippets, but not one thread devoted specifically to DR issues.

You've raised the question in relation the next 5D; if it isn't hijacking the thread I'd like to broaden the discussion to the whole issue of DR in digital photography. Is this a historico-political problem? Are we in this pickle because we haven't made our needs clear to the manufacturers? There are ten demands for better high ISO performance, more resolution, or faster autofocus for every complaint about limited dynamic range. We've seen major advances in all those other areas. If Canon thought we were really bothered by the limited DR of their products would we see that two-stop DR improvement in the 5d4?

Maybe it's a specialized problem, primarily of concern to landscape photographers? I can see where DR might not be as big an issue for studio photographers, where the photographer and not nature is in charge of the light as well as the composition. Do wildlife shooters miss the dynamic range of film?

Or is this a technical problem? Digital photography has made amazing strides over film in areas like resolution, maybe even color rendition. Modern high ISO performance makes 20th century efforts to "push" film look like something out of the bronze age. On the other hand, perhaps DR is particularly daunting challenge for CMOS technology. Would you have to give up something else to realize a significant gain in DR? Then too, is there a technical reason why Canon lags behind other manufacturers in this area?

All of this leads back to your question of whether the next few generations of digital cameras hold any significant promise for improvement in DR. Do we just have to get in Canon's face, or are we condemned to wait for an as-yet undreamt-of technology? It would be interesting to hear on these issues from people knowledgeable about the historic and technical aspects of photography.


 

[jrista]

@Curmudgeon:

It might take some real doing to get more than 14 stops of DR...that would mean truly pushing into the realm of full 16-bit RAW, and thats not an easy feat...it certainly wouldn't be cheap. However, we already have a load of "true" 14-bit sensors on the market from Sony, used in both Sony cameras as well as Nikon cameras. They are achieving over 13.8 stops of dynamic range. Given that, its obviously possible to achieve close to the full 14 stops that should be possible with 14-bit sensors and image processing chips.

Why Canon is stuck at 11.7 stops at most is really mind-boggling to me, especially given how much they tout their full precision 14-bit image processing pipeline in their DIGIC 4 and new DIGIC 5/5+ chips. Slapping a number on the side of a chip is rather meaningless if you really aren't utilizing the additional capabilities and headroom, and over the last couple of years, Canon has consistently come in second-place to Nikon when it comes to dynamic range and color depth (which really means color reproduction accuracy...and while the difference is not huge (rather largely imperceptible to the human eye in rough comparisons), it is meaningful...particularly to those who require accurate color reproduction.)

Its certainly possible to achieve 13.8 stops of DR and over 23 bits of color sensitivity/accuracy with a current-generation sensor. There is nothing to indicate Canon couldn't achieve the same, and if their new sensors in the likes of the 1D X and 5D III don't show any marked improvement once DXO tests come out...then Canon is making a choice not to improve on those fronts (to the bafflement of their customers, for sure!)

[PinkFish]

Despite the fact I really prefer the ergonomics of Canon bodies (I always think that modifying Nikon settings while in the field, is a pain in the butt), I feel more and more disappointed by Canon "politics". Body after body, the upgrades are either a small increase of features, or an attempt to catch up the competition (e.g AF, wireless flash). But no real improvement on DR.
Alas, at least in French forums, most of photogs praise the high ISO of 5DII, but few lament its poor DR. Have the last Canon landscapers fled to the competition? Or is it simply a denial ? "Au royaume des aveugles, les borgnes sont rois !"

[dtaylor]

jrista - DxO DR tests are highly inaccurate. You can disprove them yourself by simply buying and shooting a Stouffer transmission step wedge.

I don't know of any Nikon or Sony sensor which can capture 13.8 stops on a transmission step wedge. Canon's 18 MP crop sensor yields about 10 stops, while Nikon's 16 MP crop sensor yields about 11. The 5D2 is good for about 11 stops, as are Nikon's FF sensors.

In all of the above cases, if you're willing to dual process and merge your RAW file, and accept some additional shadow noise, you can probably squeeze 1 more stop out of the file, at least with Adobe's converter.

I don't see any great difference between Canon and Nikon bodies in this regard. On crop Nikon has a slight edge because their 16 MP sensor is one year newer than Canon's 18 MP sensor. I'll be curious to see what the new generation of sensors can do. But I think we're a couple generations away from a sensor which can handle, in a single exposure, landscapes with a wide scene brightness range without filters or HDR. Even color portrait films generally can't handle those scenes, and B&W only could with specialized processing and printing.

While I'm on the topic, I also consider DxO's color tests to be meaningless. Color differences between modern DSLRs are one or more orders of magnitude less than differences introduced by RAW converters, post processing, monitors, printers, and even paper choices. Sensor color differences are, with rare exception in narrow applications, effectively invisible to humans.

[LetTheRightLensIn]

jrista - DxO DR tests are highly inaccurate. You can disprove them yourself by simply buying and shooting a Stouffer transmission step wedge.

I don't know of any Nikon or Sony sensor which can capture 13.8 stops on a transmission step wedge. Canon's 18 MP crop sensor yields about 10 stops, while Nikon's 16 MP crop sensor yields about 11. The 5D2 is good for about 11 stops, as are Nikon's FF sensors.

In all of the above cases, if you're willing to dual process and merge your RAW file, and accept some additional shadow noise, you can probably squeeze 1 more stop out of the file, at least with Adobe's converter.

I don't see any great difference between Canon and Nikon bodies in this regard. On crop Nikon has a slight edge because their 16 MP sensor is one year newer than Canon's 18 MP sensor. I'll be curious to see what the new generation of sensors can do. But I think we're a couple generations away from a sensor which can handle, in a single exposure, landscapes with a wide scene brightness range without filters or HDR. Even color portrait films generally can't handle those scenes, and B&W only could with specialized processing and printing.

While I'm on the topic, I also consider DxO's color tests to be meaningless. Color differences between modern DSLRs are one or more orders of magnitude less than differences introduced by RAW converters, post processing, monitors, printers, and even paper choices. Sensor color differences are, with rare exception in narrow applications, effectively invisible to humans.

Have you ever manipulated a low ISO D3x or D700 file and then gone an manipulated one from Canon? The difference is every bit as real as DxO says it is.
Hopefully Canon can finally match it and it's not something that is only possible with something needing a Sony patent (they use column ADC on their sensors that have the amazing low ISO DR) and/or that Canon can do what is needed with their current fab.

[Curmudgeon]

Thanks for the info, jrista. I have to say I was hoping your thread would crystallize feelings and unleash a torrent of pent-up frustration among the Canon faithful.

Guess not. This thread is in its second day; it's gotten billing on the masthead, and there are a total of six responses. Meanwhile, the thread speculating about other features of the 5D3 is probably going on 40 pages.

And you know, I'd forgotten about all the fuss Canon made about  the 14-bit rendering of the 5D2 and what a big difference that was going to make. It didn't--and the preliminary talk from Canon about the improved jpeg performance of the IDX makes me worried that they may be planning to finesse the issue again in the coming models.

 

[te4o]

I'd like to join your discontent with the DR of the current Canon line. The well praised 5dII is just shocking when pushed a bit more energetically in PP. I read here about QE and SNRs etc and it is obvious that it has an old-fashioned sensor whose attraction in 2008 was the resolution. But the resolution is lost in the noisy dark areas. I hope very much as you do that the new generation of sensors will deal better with the DR of our subjects - there was a mention by someone who used the 1DX on one of the stands and got the feeling of shooting HDR - so there is hope. The question is whether the 5D line will take the full advantage of these new sensors or just part of it to make the 1D series still attractive. Who knows...
One small advantage of the rumored 6.9 FPS is that it makes HDR on windy days easier And the 4 stop AEB adds to it. But as you said, there is no substitute for a single shot of a good quality high IQ high DR. I am happy to pay for it but I am not happy to pay for 36 miserably performing MP giving me sharpness I don't need! Because it is NOT THE SHARPNESS which catches our attention, more it is the quality of light capture in DR and colors (next to subjects, composition, rarity etc etc all shooters' components)

[Mt Spokane Photography]

You don't need to do the Stoffer Wedge Test, its been done many times.  You can see the usable DR range of competiting camera bodies here:

5D MK II         8.4 EV
Nikon D700    7.8 EV

http://www.dpreview.com/reviews/CanonEOS5DMarkII/page25.asp

You can get more by manipulating RAW files, but the images tend to look flat and ugly.

I'd also like more DR, but, so far, no one seems to match film.

[DianeK]

My 60D is my first digital camera.  I was an active photographer 20 years ago mostly shooting slides, then completely dropped out of the hobby.  My re-entry into the hobby has been very disappointing...I still had fond memories of Kodachrome 64.  Seems now I am shooting a lot of AEB triads and spending considerable time in PP to increase dynamic range.  It would be very nice to not have to do that.
Diane

[jrista]

You don't need to do the Stoffer Wedge Test, its been done many times.  You can see the usable DR range of competiting camera bodies here:

5D MK II         8.4 EV
Nikon D700    7.8 EV

http://www.dpreview.com/reviews/CanonEOS5DMarkII/page25.asp

Just to make sure everyone understands the 8.4 annd 7.8 EV's from above...those are attenuated results (i.e. has a picture style...or more accurately a tone curve...applied) based on JPEG images. If you scroll down the same page linked in Spokanes post, under the RAW headroom area, it mentioned that RAW images get additional DR...and in the case of the Canon 5D II, its about 10.3 stops. Conversely, the D700 achieves 11.6 stops when tested with RAW. The dpreview tests are different (and less precise, although probably more realistic) than DXO's tests.

I would say its difficult to directly compare results from dpreview as a measure of low-level camera capabilities, but a great way to measure real-world capabilities. Conversely, I would say DXO is a poor way to compare cameras from a real-world standpoint, but it is an accurate way to compare minor low-level differences in hardware capabilities.

Regardless, both DXO and dpreview indicate that, when using RAW, Nikon (and Sony, for that matter) cameras get better dynamic range than Canon...by over a stop in most cases, and as much as two for the best Sony sensors.

You can get more by manipulating RAW files, but the images tend to look flat and ugly.

I'd also like more DR, but, so far, no one seems to match film.

Thats not entirely accurate. You have more headroom with a RAW image. They are still attenuated with a tone curve, so contrast improves over the unprocessed native sensor data (which you can generally only see with open-source RAW processors...ACR/Lightroom and Aperture all apply non-linear base processing to all RAW images). You have a good 1.5 to 2 additional stops of dynamic range headroom with RAW, though, so you are less likely to blow out highlights when shooting, say, a sunset, as you would be if you had two stops less DR. I'd rather have that 10.3 stops than the 8.4, and having 11.6 stops of real-world DR would be bliss. The extra headroom means making photographs is "safer" from hitting that hard limit that digital sensors have on the highlight end of things. In film, highlights "blow out" slowly (its analog after all), there is no hard digital limit, which is why film is still touted as having better highlight flexibility than digital.

[LetTheRightLensIn]

You don't need to do the Stoffer Wedge Test, its been done many times.  You can see the usable DR range of competiting camera bodies here:

5D MK II         8.4 EV
Nikon D700    7.8 EV

http://www.dpreview.com/reviews/CanonEOS5DMarkII/page25.asp

You can get more by manipulating RAW files, but the images tend to look flat and ugly.

I'd also like more DR, but, so far, no one seems to match film.

DPR test is ridiculous, it depends on nothing other than what tone curve is in the jpg setting.

[mkln]

I have a hard time believing DXO and some of their ratings, often times they look so unrealistic I could S___ in my pants.
for example the rating they give the 70-200 II is low, much lower to what they give to 85L. truth is, I have both and find 70-200 as good as 85L (of course they measure max performance throughout the aperture and zoom range, and I can't agree with them on that)

HOWEVER, their sensor ratings seem to be accurate.
I have a 5d2 and a NEX 5N. they say the 5d2 should have less DR than 5N, and indeed that's what I perceive.

When I bought the 5N I liked its size and its being lightweight. Then I took some pics and I was amazed at how it didn't blow up the highlights. only later did I check on DXO to see what kind of DR they measured on it.
turns out it has 2/3 of a stop more than the 5d2. and I really think that's the case.
now I enjoyed the 5N shots so much that I will not upgrade to 5d3 (not at that price) unless it shows significant DR improvement over 5d2.
and by significant I mean at least 1 stop. Sony has proven it possible... why can't Canon?

I really don't understand all that fuss about FPS. if you want fps just buy a used 1D3, or a 1D4. I bet their prices are around what the 5d3 will be (new). and they have the "extra reach" that sports photogs seem to love.

Also I don't understand why people look at the D800 and only see 36mp and say "omg it's too much it's going to be blurry/poor high iso/other" when it will have the same pixel density as my 5N and:
- great DR
- my 5N does not produce blurry pics AT ALL, required shutter speed is not significantly different from what I use with the 5d2
- high iso on my 5N are great. in quantity terms not as good as the 5d2 (FF vs. APSC, that's a given), but in quality terms it's much more manageable (no banding). If I were to extend the 5N sensor to full frame (that's a D800) and to resample the pic to 21mp, I bet the result would be much better than the 5d2.

so yeah, D800 has high mp, but it also has an amazing sensor that excels in everything, DR included.
and yes, canon fanboys are blind to DR, they only want FPS and AF (btw, good AF is the very least we could expect from a $3500 camera, duh!!)

[Curmudgeon]

Hopefully Canon can finally match it and it's not something that is only possible with something needing a Sony patent (they use column ADC on their sensors that have the amazing low ISO DR) and/or that Canon can do what is needed with their current fab.

When I consider the relatively slow progress Canon has made in DR, I also sometimes wonder if Sony has some recent patents that Canon is having difficulty circumventing.  Canon is now competing with some heavy hitters in the electronics industry. LTRLI, can you expand a little on what column ADC is?

There were also some older technologies that had some success in expanding DR. As I recall, a few years back Fuji was building DSLRs that featured ganged pairs of high and low sensitivity pixels. Again (if memory serves) Fuji was in the unfortunate position of having to retrofit their technology into Nikon D200 bodies, but apparently the cameras actually did have more latitude than the CCD sensors of the time and enjoyed some popularity among wedding photographers. Anybody have any experience with those cameras or know why that technology faded from the scene?

[jrista]

When I consider the relatively slow progress Canon has made in DR, I also sometimes wonder if Sony has some recent patents that Canon is having difficulty circumventing.  Canon is now competing with some heavy hitters in the electronics industry. LTRLI, can you expand a little on what column ADC is?

Its actually called Column-Parallel ADC. Normally, a single ADC (analog-to-digital converter) is used to convert all of the analog signal read from a sensor into a digital counterpart. Sony has a patent (I believe, anyway) for a CMOS design that uses multiple ADC's in parallel for each column of pixels, theoretically improving readout throughput considerably.

I am not exactly certain how Canon's 1D X sensor works from an actual hardware layout level, but it does have 16 parallel primary read channels, which in turn are split into two groups of 8, each group of which is handled by a dedicated Digic5+ processor. I would imagine that would require at least as many ADC's, one per read channel. I am not sure if Canon's approach reduces read noise levels to the same low level as Sony's, but if they are really able to reach a native ISO 51,200 with the same noise level as present at ISO 6400 before, it must have improved it enough.

I can't say whether the 1D X has column-parallel conversion on each and every pixel column, but seeing as Sony hasn't yet achieved the full theoretical 28fps @ 18mp that should be possible, I figure their C/P ADC isn't quite as good in real life as it sounds on paper.

There were also some older technologies that had some success in expanding DR. As I recall, a few years back Fuji was building DSLRs that featured ganged pairs of high and low sensitivity pixels. Again (if memory serves) Fuji was in the unfortunate position of having to retrofit their technology into Nikon D200 bodies, but apparently the cameras actually did have more latitude than the CCD sensors of the time and enjoyed some popularity among wedding photographers. Anybody have any experience with those cameras or know why that technology faded from the scene?

Fuji's diagonal grid with extra luminance-only sensels did pretty much the same thing that microlenses does on Canon sensors. It helps utilize otherwise wasted space to gather light that would have been reflected by wiring or simply unused wafer surface. With the advent of gapless microlensing, there is very little sensor surface that can't be used to gather photos and direct them to a photowell. It would actually be fairly difficult these days to develop a sensor that had luminance only pixels without taking a step back from where we are with microlensing. Front-illuminated sensors still have obstructions in the light path to the photodiode itself, which can either absorb photons and convert it to heat or noise, or reflect them back out of the sensor. A back-illuminated sensor would reduce the chances of absorbing or reflecting photons before they are captured by the photodiode. Beyond back illuminating...I think the only real improvement would be to move to a layered design like Foveon. Thats the only major remaining drawback of the bayer sensor design, however at the densities they are achieving today vs. the highest density Foveon sensors, the difference is largely moot (an 18mp APS-C/45mp FF has enough density that for realistic presentation formats...moving to a lower-resolution Foveon sensor wouldn't offer any realistic benefit outside of potential for improved DR, reduced color moire, and greater color sensitivity.)