A bit confussed

[FloridaBoy]

I own the Canon 5D Mk II. I have tested the Dynamic Total Tonal Range to be 9 stops. It has a 14bit A/D converter that is capable of recording 16,384 tonal values, over 9 stops. The A/D converter info is relative to info on this site.

14Bit A/D Converter specs------------------------Dynamic Tonal Range of the 5D Mk II

Stop9-Brightest tones........8192 Tonal values-----White without color cast and detail
Stop8-Next Brightest tones.4096 Tonal values-----Tonal shift without detail
Stop7-Darkest Bright tones.2048 Tonal values-----Tonal shift with detail
Stop6-Upper middle tones...1024 Tonal values-----Tonal shift with detail
Stop5-Middle tones.............512 Tonal values-----Tonal shift with detail
Stop4-Lower middle tones....256 Tonal values-----Tonal shift with detail
Stop3-Lightest Dark tones....128 Tonal values-----Tonal shift with detail
Stop2-Next Darkest tones......64 Tonal values-----Tonal shift without detail
Stop1-Darkest tones.............32 Tonal values-----Black without color cast and detail

My confusion rests on the comparison of the A/D specs relative to the cameras capture D/R. My camera specs show me that I have detail and tonal shift in 5 stops (stops 3-7), yet my converter applies the most tonal values to the top two stops. Those stops have no detail nor tonal shift. Can someone explain to me why this is not correct or really is productive? I am just looking to understand.

[ajohnw]

Are you looking at the images on a PC screen? Are you looking at camera jpg's? Could also do with some idea how you are doing this test.

John
-

[dubaiphil]

but does it take nice pictures???

[Manfred M]

First of all , I suspect your test methodology is where I would start looking for issues, and as you haven't described what you've done, it's impossible to comment and critique it.

Secondly, I think you are confusing how 14-bit encoding and the actual sensor sensitivity corrolatation. Reputable test lab results suggest that the dynamic range of the sensor in the camera is 11.7eV; http://www.dxomark.com/Cameras/Canon/EOS-5D-Mark-III.

While the A/D converter limits the number of discrete values to 2^14 (0r 16384) per channel, all this shows is that whatever the sensor is capable of outputting is bucketed into this number of discrete values. The 11.7eV on the 5D MkIII are distributed into 16384 indivudual "buckets", just like the 14.4eV on my D800 are done the same way. The dynamic range of the sensor is what it is, all the A/D converter and bit depth control is the discrete distrubution of data into the individual "buckets".

The same sensors could be encoded into 12-bit or 8-bit data (in fact that is exactly what happens when you convert to a jpeg), yet the sensor dynamic range will still be what it is. I hope this makes sense.

[FloridaBoy]

I feel that I am stating a Norm for all sensors. Regardless of number of stops, the sensor is going to give you a White without detail or tonal shift in the brightest stop. It is going to give you a Black without detail or tonal shift in the darkest stop. For the stop next to the white and black, you will have a tonal shift but No detail. Only the middle stops will give you detail, and tonality shift. The nature of the beast now and in some ways the days of film.

As for factory/lab tests. I would suggest that we all test our own cameras. In a world of mass production and +/- acceptable values, I believe that most will be surprised. I know that I was. I make exposures in the real world and not a lab.

As for my test methodology. I started this game in 2010 with Lee Varis book "Mastering Exposure and the Zone System for Digital Photographers" He laid out the ground work, on camera dynamic range for me. Two other, very helpful individuals, filled in the blanks. I am comfortable that my 5D Mk II is a nine stop total tonal dynamic range camera and Not 11-12 stops as the "labs" have stated. The reason that I pursued the D/R avenue was; because the lack of D/R in my digital camera was visually less than my days of color negative film. The film of the '80's was stated as being 11-12 stops. Kodak Ektar 25 color negative film.

My question is: With sensors that provide "only" white and black in the brightest and darkest stop, why is the A/D converter putting one half of the total tonal values in a white/brightest stop that has no detail or tonality? If this is a fact, then exposing to the right, may come with thoughts of exposure shift, for those practicing the zone system.

[FloridaBoy]

{While the A/D converter limits the number of discrete values to 2^14 (0r 16384) per channel, all this shows is that whatever the sensor is capable of outputting is bucketed into this number of discrete values. The 11.7eV on the 5D MkIII are distributed into 16384 indivudual "buckets", just like the 14.4eV on my D800 are done the same way. The dynamic range of the sensor is what it is, all the A/D converter and bit depth control is the discrete distrubution of data into the individual "buckets".}

This statement does not follow what is stated on this site. Only the brightest stop gets 8192 in that bucket. The remaining stops get divided by 1/2 total tonal values left per stop. The second stop would only get 4096 and so on.

[ajohnw]

Maybe this will help the OP but a bit of an explanation is in order especially as there is a lot of complete and utter bunkum about on this subject.

This pages shows your camera's dynamic range in graphs against actual EV values = stops. The one to look at first is the jpg contrast curve plot with contrast at -4, 0 and +4. Notice that the slope of the line changes and the the end curl. Visually in an image the 0 contrast one will give realistic tones where the line is more or less straight. Notice that the +4 increased contrast line is steeper and that the -4 has less slope. Basically the slope of the line indicates contrast and 0 corresponds to what could be called normal as far as an image is concerned.

Where these lines curl over the slope gets less so the contrast must as well. In other words it wont look natural.

Further down there is a section called ISO setting and dynamic range. This reduces as the ISO goes up due to noise. The noise mostly causes problems at the darker end of things which is why these figures change far more than the highlight end. The ends of these lines still curve. The dynamic ranges mentioned are probably just achievable in realistic looking image terms but only via post processing. As they come the dark areas will lack contrast as you have noticed. You may not have noticed that the bright end has the same problem as well. The effect is most noticeable in darker regions.

Then comes raw processed images using Adobe Raw.

Even further down there is a section called raw headroom - Canon are famous for leaving a fair amount of that. This is where a certain amount of specmanship starts creeping in. You can compare the slope of the lines against the jpg curve as that is shown as well. That represents normal contrast where it is straight. The others curves are interesting. Adobe default gives the same results as jpg. Adobe auto corrects the highlight contrast levels and gives a good contrast slope from mid tones up and reduces contrast below that.. Adobe best is probably a bit of a joke. The black end has been lifted a lot which will increase any noise as well and most of the slope of the line is well below normal contrast levels.

Where things get confusing in this area is when the shot is viewed on a screen. A decent monitor has a 5 ot 6 EV capability and the camera has more. All of the graphs mentioned above are against how they would appear on a monitor and indicate how much PP will be needed to make them look natural. If the cameras full dynamic range is used in a shot the dark end will have to have it's contrast boosted so that it looks natural - a fill light is the easy option. It's likely to need brightening as well as monitors do not work very well at the very dark end because of back lighting - black can't be black. As you seem to have noticed there are limitation on just how much the dark end can improved due to colour changes/noise. The more EV from the camera that has to be squeezed in the more difficult the PP will get. The bright end can have similar problems and also can only be so bright because of clipping at the brightest level available. This can even mean dimming it so that the contrast can be changed. The adobe auto mentioned looks a very attractive option but in practice much will depend on the shot.

http://www.dpreview.com/reviews/canoneos5dmarkii/25

There is another aspect to this area as well. If our eyes look at say 7 EV steps of light we will see 7 equal brightness changes if they are bright enough - more or less. We see logarithmically. The monitor and the screen work linearly so say some stop only has a range of 8 we might not be able to detect light level changes from say 4 to 8 because there isn't sufficient light change for us to detect. If one the other hand some EV had a range of 16 we might see a change from 8 to 16 and maybe gradations in between. In a nut shell this is why a 14 bit A/D can't really produce a 14 EV image that is of any use to us.

This should help if you spend some time thinking about it. It's a fact that despite claims in reviews dynamic ranges haven't improved all that much for some time. The improvements are mostly down to introducing more curve at the ends of the lines, Looks good and PP techniques have improved as well so can be of some use - maybe. It all depends on the shot.

Out of time so hope not too many odd typo's. I make some really weird ones at times.

John
-

[Manfred M]

I feel that the explanation is a bit dated and at the time the article was written it may have indeed been representative of the technology at the time it was written. The math is obviously still the correct, it’s just the outcome of applying it that will have changed.

The stairs and step height analogy is a good way of looking at the issue (the steps are the "buckets" in my terminology). A sensor with a higher dynamic range would still be able to discern shades of gray where one with a lower dynamic range would be capturing these data points as pure white and pure black. What the bit depth of the A/D converter does control is the contrast range. Using the stair analogy, the steps would be taller with the higher dynamic range sensor, but the number of steps would be the same. So frankly, as long as we have a sensor-A/D converter pairing that exceeds the16384 discrete resolvable values; any additional improvements are meaningless, as we won’t be able to distinguish the differences between the discrete levels. Nicely said, I can’t tell the difference in output I see from a 5D Mk III versus a D800

Frankly, other than in some editing work, it really makes no practical difference. If I remember correctly, the dynamic range of a high end computer screen is around 5 or 6 stops and a print is around 4 stops, so whether the camera can capture 11.7 stops or 14.4 stops, it really does not impact any real world outcomes.

[Manfred M]

Good explanation John.

I do know that some camera manufacturers use the "headroom" in the top and bottom end (the so-called shoulder and knee parts of the curve) to drive additional separation at the ends of the curve to provide additional detail in the highlights and shadow detail for jpeg and video (compressed output as well). From a technical standpoint it has advantages, but the it's all too easy to rely on it to much and end up blowing out the highlights.

[ajohnw]

They have been bringing the dark end "headroom" in for a long time. It's a bit disgusting really as some one could look at a jpg curve and say I can get 10 stops. The problem is that it's not going to be easy to get that dark end looking natural. Lowering the contrast of the bright end isn't so bad as stops up there have bigger numbers so gradation seems to is apparent and crunching it up isn't so noticeable.

Nikon have been known to over expose to bring the blacks up a bit. Now D-Light. Olympus don't leave much headroom but the metering reflects this really. They also have a jpg curve along the lines of the acr auto in the review. They are all playing at similar things and they can make the numbers look better than they really are.

One interesting aspect is that few dpreviews are as thorough as the 5D MkII and certain tests come and go. I'd guess that they want to continue to receive free cameras for testing. One aspect is that they do generally mention problems if carefully read.

John
-

[FloridaBoy]

They have been bringing the dark end "headroom" in for a long time. It's a bit disgusting really as some one could look at a jpg curve and say I can get 10 stops. The problem is that it's not going to be easy to get that dark end looking natural. Lowering the contrast of the bright end isn't so bad as stops up there have bigger numbers so gradation seems to is apparent and crunching it up isn't so noticeable.

Nikon have been known to over expose to bring the blacks up a bit. Now D-Light. Olympus don't leave much headroom but the metering reflects this really. They also have a jpg curve along the lines of the acr auto in the review. They are all playing at similar things and they can make the numbers look better than they really are.

One interesting aspect is that few dpreviews are as thorough as the 5D MkII and certain tests come and go. I'd guess that they want to continue to receive free cameras for testing. One aspect is that they do generally mention problems if carefully read.

John
-

[FloridaBoy]

Sorry to say it, but little has been accomplished with this post or it's replies. I brought my question here because the info at the top left of this post was generated by this Cambridge website. Yet replies ignore that and take me to dpreview, jpgs, and talk that is about Total Tonal Dynamic Range. None of it addresses Tonal Dynamic Range nor Detail Dynamic Range nor the real raw world. As well no one has said that the original info was true, false, or irrelevant. The dpreview info is nothing new to me and if it answered my stated question, I would not have posted here. Your best statement was; dpreview kept the 5D Mk II up front so they could obtain future Canon cameras to preview. That says more than any words spoken here.

I shoot Raw only, tested my camera in Raw and evaluated in ACR set to neutral. I have not entertained Jpg since 2007.

I appreciate the dance, but your lyrics are not of interest me.

[pnodrog]

The A/D converter specs have nothing to do directly with the tonal range.... It just defines the number of different voltage levels it can resolve from the output of the sensors amplifier. The fact that the tonal range is logarithmic (base 2) tends to confuse our thinking. If the A/D was measuring the output from say a linear temperature sensor that could measure up to 1638.4 degrees we would know if it was a perfect world that we could resolve to 0.1deg. (Actually I think a humans response to temperature is also non linear but we seem to get far less confused about temperature measurement)

It is convenient because of the binary output from the A/D to assume that it relates directly to the tonal range but that is not the A/D's function.

[ajohnw]

Sorry to say it, but little has been accomplished with this post or it's replies. I brought my question here because the info at the top left of this post was generated by this Cambridge website. Yet replies ignore that and take me to dpreview, jpgs, and talk that is about Total Tonal Dynamic Range. None of it addresses Tonal Dynamic Range nor Detail Dynamic Range nor the real raw world. As well no one has said that the original info was true, false, or irrelevant. The dpreview info is nothing new to me and if it answered my stated question, I would not have posted here. Your best statement was; dpreview kept the 5D Mk II up front so they could obtain future Canon cameras to preview. That says more than any words spoken here.

I shoot Raw only, tested my camera in Raw and evaluated in ACR set to neutral. I have not entertained Jpg since 2007.

I appreciate the dance, but your lyrics are not of interest me.

I didn't say dpreview kept the 5d mkII up front at all. I said it was a very thorough review. As to the rest I sometimes wonder why other reviews are less thorough.

As I said it needs some thought and the curves do explain the problems you have noticed or read about. I'm truly glad you understand it all and don't need the reply. I should have added a bit more stressing what noise does to the numbers. Chroma noise exists at all light levels and has more effect at the dark end for reasons you are clearly already aware of - the numbers that record the light levels are smaller. Just "brightening" is no good. Contrast has to be increased. That amplifies the changes in light level and any noise.

Really rather then read about it all it's better to take some shots and find out for yourself. For instance I know that if I have a high dynamic range scene which basically means there will be areas that are much darker than they appeared to be when I looked at the scene I understand why. I also know that in my terms I can recover a certain amount of detail from a camera jpg without problem. Also that I can recover more from a raw file but how much depends on the camera and it's settings. Same at the bright end.

Dpreview - I would estimate that there is some sense in what they quote on usable range in this particular review but it may take a whole lot of fiddly PP. The more radical figures from raw depend on a number of things but are worth trying on shots just in case they leave less to do. More often a manual curve is more appropriate. The numbers they give on all cameras do give some idea what they actually can do. Simple fact. The manufacturers will stuff as much as they realistically can in a jpg. Of late as cameras evolve many will try and do something about dark end problems in jpg's as well. That aspect can give people and idea of capabilities in that direction. As you understand it all you should find the site very useful.

John
-

[Manfred M]

Actually, I’m still not quite sure as to what you are asking.

L Paul’s comments on what A/D convertors do is spot on. They are very simple devices and assign a numerical value to a contiguous range of voltages. A 14-bit converter does nothing more than to take a specific input voltage range and assigning it one of 16384 discrete values.

As for factory/lab tests. I would suggest that we all test our own cameras. In a world of mass production and +/- acceptable values, I believe that most will be surprised. I know that I was. I make exposures in the real world and not a lab.

Frankly, if I got test results that are significantly different from the results of a reputable testing lab, like DxO labs, I would question my own test methodology, rather than assuming that the published results are incorrect. I suspect that your underlying conclusions are based on questionable test results, but as I stated in my original post, unless I understand your test methodology, I can’t comment on the validity of those results.

An interesting assumption, but as someone who has spent many years in technical role in manufacturing and production environments, I would have to suggest I would like you to provide some backup to your statement. The micro-electronics components manufacturing process is remarkably accurate and repeatable. The complexity of these products means that they are manufactured to exacting standards using a highly repeatable process; with limited exceptions, failure tends to mean the component is scrapped.

While I will be the first to suggest that many photographic tests to not represent real world results, establishing the maximum dynamic range of a particular sensor would not fall into this category.

As for my test methodology. I started this game in 2010 with Lee Varis book "Mastering Exposure and the Zone System for Digital Photographers" He laid out the ground work, on camera dynamic range for me. Two other, very helpful individuals, filled in the blanks.

I’m afraid I have not read Lee’s book, but I did have a quick look at his CV. Lee is a graphic artist, not an engineer, so I would tend to be a bit suspicious of his technical credentials. I know quite a few professional photographers (a number of them are photography instructors at the local community college). While I fully respect their photographic knowledge and technical expertise when it comes to photographic fundamentals, I’ve certainly heard them say things about digital technology that do not stand up to technical scrutiny. Regardless, your response does not describe any aspect of your test methodology.

[george013]

@FloridaBoy,
Your confusion starts with a wrong interpretation of 'bitdepth' or 'tonal range'. You can't split that range of values in stops. You can divide it in smaller parts but not in stops. I know that this can be difficult to understand but without understanding this you won't see the light( I can't find the smileys)
George

[ajohnw]

Seems to be a waste of time but reposting part of the original post

Stop9-Brightest tones........8192 Tonal values-----White without color cast and detail
Stop8-Next Brightest tones.4096 Tonal values-----Tonal shift without detail
Stop7-Darkest Bright tones.2048 Tonal values-----Tonal shift with detail
Stop6-Upper middle tones...1024 Tonal values-----Tonal shift with detail
Stop5-Middle tones.............512 Tonal values-----Tonal shift with detail
Stop4-Lower middle tones....256 Tonal values-----Tonal shift with detail
Stop3-Lightest Dark tones....128 Tonal values-----Tonal shift with detail
Stop2-Next Darkest tones......64 Tonal values-----Tonal shift without detail
Stop1-Darkest tones.............32 Tonal values-----Black without color cast and detail

Why doesn't stop 8 show detail? It should. What will interfere with that detail is the curves I went through. They will generally have even more effect on stop 9. Detail is contrast - no contrast no detail - reduced contrast less pronounced detail. The other point of course is that increasing the A/D bit depth doesn't do anything other than multiply the numbers by factors of 2. 4 if moving from 12 to 14 so great any noise produces 4 times bigger numbers too. The only region it may be of any use in is the lower tones given that noise levels are usable. The bigger number means that contrast can be stretched more precisely to make the tones look better in the final 8 bit space.

A jpg is just a raw conversion using a standardised tone curve that as it happens is generally very close to adobe standard. The advantage of raw is that it retains the full data from the camera sensor. It's been truncated to 8 bit in a jpg and probably had a cetain amount of smoothing done to it. The other advantage of raw is that the tone curve used for the conversion can be changed. That can make a considerable difference to what comes out of raw. Most people seem to use adobe standard or one of landscape etc settings. Just means in real terms that the immediate results will be very similar to camera jpg's taken with the same curves. Things aren't that simple in respect to Nikon. They generate a curve for every shot.

If some one wants to actually test all light levels up to the top of stop 9 it's not difficult to do. Fire up an editor and generate a white area, point the camera at it so that it fills the view, out of focus, set up to take an exposure via the spot meter and take it. That's the staring point, mid grey. Then take a series of exposures with different EV compensations. Ideally these all want to be at the same shutter speed. It should be possible to do this. If there aren't enough aperture stops on the lens go back to the original settings for mid grey and then dim or brighten the screen by so many EV. Ideally a high precision light meter should be used to do this or go out and buy a precision stop wedge such as dpreview use. Stick a light behind it of the right type and then calibrate mid grey. It should be possible to get 1/3 ev step shots like this with most cameras and the camera metering should be stable at least and some error in that can be accounted for. Using 1/3 stops should indicate if tonal variation is possible in a particular stop. They can be PP'd too. That will show scope for recovery into the eventual colour space which is what actually matters.

The mid grey shot should measure at 18% grey =18% on the way to pure white visually. On the other hand there seems to be 2 standards kicking around on this aspect so some web searching might be a good idea. The ideal way to measure that is to have software that reads all of the numbers out directly from the raw file. If the raw file is developed with a linear conversion curve and no tone mapping at all the pixels should sit at 18% on a pc screen but unfortunately they emit light. Some seem to think mid grey is represented by 127 on a PC screen. Pass.

There are all sorts of holes in this but the changes between the 1/3 EV steps will show if a particular stop can show detail and how well. Frankly though people have done this sort of thing I feel it's easier to look at and accept independently measured tone curves because then it doesn't matter a hoot how many bits of precision the A/D has, it just needs to be more than 8 which is why compacts use 10. 8 wouldn't generate decent jpg's as any A/D reading can be +/- 1 bit out. At some levels reading say an 8 as 7 or 9 could make a significant difference to the image.

If memory serves me correctly cameras can vary in terms of mid grey to either 13 or 18%. Looking quickly this person does seem to know what they are talking about.

http://www.ryanewalters.com/Blog/blo...83675371023276

14 EV dynamic range mentioned. Really? It seems to me that there are some rather gullible people about if they believe it has any bearing on practicalities.

John
-

[Colin Southern]

I own the Canon 5D Mk II. I have tested the Dynamic Total Tonal Range to be 9 stops. It has a 14bit A/D converter that is capable of recording 16,384 tonal values, over 9 stops.

Just a little side-note here:

You appear to be confusing dynamic range (measured in stops) and ADC resolution (NOT measured in stops); the two aren't particularly related.

The dynamic range is "the height of the staircase" whereas the ADC resolution is "the number of steps on that staircase". You could have a 4 bit ADC on a camera with a 20 stop DR ... it would simply be "big steps" between levels.

[oldgreygary]

I think my confusion is around that if detail at this level is worrying you, How do you actually manage to take any pictures?

Gary

[ajohnw]

Just a little side-note here:

You appear to be confusing dynamic range (measured in stops) and ADC resolution (NOT measured in stops); the two aren't particularly related.

The dynamic range is "the height of the staircase" whereas the ADC resolution is "the number of steps on that staircase". You could have a 4 bit ADC on a camera with a 20 stop DR ... it would simply be "big steps" between levels.

I think the other confusion comes from the fact that these 9 stops can not be viewed without curves or to really see them individual shots. All good for reviewers and fudging huge dynamic ranges such as the adobe best in the dpreview link. Maybe someday some one will do a zone system for a PC screen. If we ever get 10 bit colour no doubt people might think more dynamic range but all it can really offer is more colours.

John
-