kidtexas wrote:
So in this case, does the 12.7 (or 13.7 or whatever the quoted number is) stops of DR represent that many recordable stops of exposure in the scene? As in if I took a picture of a scene, and then I spot metered the darkest shadow recorded before the signal is lost in the noise, and the brightest highlight before clipping, the difference between the two spot readings would be 12.7 stops?
I'm aware of the differences between bits of AD, stops, and the fact that just because these cameras have 14 bit converters, they aren't necessarily any better off than they were with 12 bit ADs. I'm just unsure of what DXO means when they say DR. The only DR I really care about is how much of the dynamic range in the scene I can capture. From that point, its easy to map the photo onto the compressed dynamic range of paper, either in Photoshop or in the darkroom....Show more →
Yes, the quoted DR in DXO mark should be indicative of the number of recordable stops in a scene. D3x is one of the first cameras where the higher bit rate AD conversion is needed. The actual DR figure in real world conditions is probably open to debate, but the performance relative to other cams should be reliable.
Interestingly this camera reads the sensor data much slower when doing 14-bit conversions. Some people are speculating that the improved DR has as much to do with the slower and more accurate sensor readout as the sensor design itself. Frame rate drops dramatically, but who shoots landscapes at 8fps?
Hmm. Well I'd love to see some tests of real world DR done with this camera to back up this measurement... Something where a scene is constructed AND metered for like 15 stops or whatever, and then see where the camera stops recording detail.
Over a late nite curry we discussed DR and other stuff...
Apparently DR is a compromize, i.e. optimizing highest DR will give poorer performance at high ISO's and vice verse.
Looking at how the DR curves are shaped, the implementation of Canons seems to hold up better at high ISO... so it seems there is no "Jack of all trades", it comes down to physics
I'm sure some with better understaning can give us an insight on how this all come together
andrewd01 wrote:
Its nothing to do with the AD converter. Dynamic range is the ratio of the largest to smallest signal. This can be expressed in binary notation, and is done so in many articles. Since each f stop lets in twice as much light the DR from a given number of fstops is the number of stops to the power of 2, therefore it corresponds directly with binary notation.
For example two f stops of DR, the largest signal at say f16 , the smallest at say f22, the DR = 2. This value requires two binary digits to record in binary format, so the DR can be expressed as "2 bit"
Now lets consider an image with 3 stops DR (say f22, f16, and f8) f8 gathers 4 times as much light as f22, so the DR is 4. This would require 3 binary digits to represent (100). And so on.
So you see that DR can be expressed as bits, and that the number of bits required to represent the DR is the same as the number of f stops.
I would encourage you to be more open to new concepts given on these forums, you will learn more that way. Everyone has a slightly different background and often concepts that we are familiar with are presented in a slightly different way to which we are used to. Nuff said on the matter. ...Show more →
Don't be so condescending. No such thing as 13.7 bits anyway you cut it. Now if you would get off your high horse and say that DR is 13.7 stops then we will agree. Just like statistics that say the average family size is 2.7 kids - no such thing. It is meant to illustrate that the average family size is 2 or 3 kids. The equivalent for this conversation would be to say that it takes 13 or 14 bits to represent the DR. By the way - my background is as a software developer and electronic engineer. I've designed A/D's using Analog Devices cells. Worked with machine vision too. It is also impossible to talk about bits without mentioning the AD - how do you think it gets into digital form? So how many bits for the 4.7 Dmax on fiber paper? Silly isn't it? Yet Dmax is a term used for DR on paper. By your argument bits can be substituted for Dmax. So - you're right - enough said on the subject - you have a grip on the subject, your terminology is simply up for question.
kidtexas wrote:
So in this case, does the 12.7 (or 13.7 or whatever the quoted number is) stops of DR represent that many recordable stops of exposure in the scene? As in if I took a picture of a scene, and then I spot metered the darkest shadow recorded before the signal is lost in the noise, and the brightest highlight before clipping, the difference between the two spot readings would be 12.7 stops?
I'm aware of the differences between bits of AD, stops, and the fact that just because these cameras have 14 bit converters, they aren't necessarily any better off than they were with 12 bit ADs. I'm just unsure of what DXO means when they say DR. The only DR I really care about is how much of the dynamic range in the scene I can capture. From that point, its easy to map the photo onto the compressed dynamic range of paper, either in Photoshop or in the darkroom....Show more →
Just a quick note - 14 bits is always better than 12 bits if they're really being used. You have a better tonal range. you can always get black and white - it's those tones between you gotta worry about . It's been discussed on LL in their exhortations to "expose to the right". The general premise is accurate so I'll discuss that instead of the "gotchas". Since the range in bits is proportional to the stops of DR an interesting observation comes to bear - each bit represents a stop (as has been discussed and debated - my quibble is there's no such thing as a partial bit). So take a scene that is exposed to produce the complete range from white to black. White is represented then by a value of 2^14 or 16384. That first stop of exposure has a range of 2^14 - 2^13 = 8192 levels - fully half of your available tonal range is in the first stop (take away a bit to get the range of the next stop down). Some might say that fully half of your DR is in that first stop but that's a gray area on how you define DR - I prefer to say you have half of your tonal values in that first stop. Keep going and you find out that for each stop your tonal range is cut in half. By the time you get to that last stop you find out that it's either there or it isn't depending on how frisky the electrons felt or which side of the quantize time it fell on. So - do you consider there or not-there as a significant factor of the DR? Bear in mind we're talking blacks here and if I remember right it has been determined that 16 levels - 4 bits - is the minimum discernible between tonal values. To my mind that automatically cuts off 4 bits as a factor in DR as equivalent to number of bits. The 5D has a 12 bit converter with a theoretical DR of 12 stops. In the real world it has been determined that the actual DR of the 5D is somewhere between 7 and 8 stops depending on who runs the tests. This would bear out using 4 bits as a minimum discernible tonal range: 12-4 = 8. In the real world we use EV's to figure exposure. A rough figure is 16 for a sunny day. Roughly speaking you can speak of EV's and f-stops as the same. Anyone who has shot a sunny day scene with shadows knows that something's gotta give - typically the shadows or possibly you decide the sky can be blown. The DR of the camera is a sliding window on the exposure range of the real world - so call it 16 or 17 stops depending on your location. I'll leave it to you to draw your own conclusions on how accurate the quoted DR is for the Nikon and how well it will capture the scene.
Regards,
Jon
Canon's implementation may bring a slightly higher DR at the higher ISO settings but comes at the cost of higher noise than Nikon has at those same settings. When the light is so weak that you have to shoot at 3200 or 6400 just to get the shot, which are you going to be more concerned with? Dynamic range comparisons beyond about ISO 400 are not of much value because anyone who is trying for great DR will be looking for good light.
WmPat wrote:
Canon's implementation may bring a slightly higher DR at the higher ISO settings but comes at the cost of higher noise than Nikon has at those same settings. When the light is so weak that you have to shoot at 3200 or 6400 just to get the shot, which are you going to be more concerned with? Dynamic range comparisons beyond about ISO 400 are not of much value because anyone who is trying for great DR will be looking for good light.
Yeah - in practice he gain is boosted for the higher ISO - typically the shadow/midtone resulting in a compressed tonal range and higher noise = < DR. TANSTAAFL (Robert Heinlein acronym)
BubbaJon wrote:
Just a quick note - 14 bits is always better than 12 bits if they're really being used. You have a better tonal range. you can always get black and white - it's those tones between you gotta worry about .
Agreed, but in most (if not all) of the cameras on the market, it seems like they're not really being used. Maybe the D3/D3x/D700 is on the verge of making them useful. I don't know.
kidtexas wrote:
Agreed, but in most (if not all) of the cameras on the market, it seems like they're not really being used. Maybe the D3/D3x/D700 is on the verge of making them useful. I don't know.
No - I can see the difference in my 5D2 shots - much smoother in the shadows and the images take more twiddling than the 5D images. They are being used. What I was referring to was some converters are designed to "skip" the lower bits for speed - the last two stops (least significant bits) are a fixed value - so really not used.
bobbytan wrote:
1Ds IV rivaling the S2? I don't think so. It's like comparing a 35mm SLR to the Pentax 6x7 in the old film days. Apples to oranges. Size matters.
Is that why the M8 sensor is so much better than the 450D sensor? Basically I agree with you, but we should remember that better technology can overcome the size advantage.
All this measurement is fine, but I would like to see how each camera copes with highlight retention and shadow retrieval for a particular high contrast scene. The number of bits isn't so relevant - what counts is the ratio of light strength between highlights and shadows which can be captured without blowing the whites and fogging the blacks at any particular iso. In that sense you could have a 4 bit camera with 20 stops of DR. That is best illustrated with examples, not graphs.
Basically I agree with you, but we should remember that better technology can overcome the size advantage
Partly. For noise, for colour and contrast, for dynamic range, yes, for a great part. BUT: what technology won't overcome is the "sculpture effect" the "plastique" (hard for me to find the right word as a non native english speaker) of a larger format. 67 differs from 35, like the S2 differs from full frame. Effects on DOF and the "sculpture" of the image from a larger sensor are very well perceptible, just like they are perceptible between APS-C and a full frame capture. Indeed, size matters, even when technology does compensate partly: "there is no substitute for sensor size!" Have (full frame) EOSfun
Now I never like to argue with eosfun, but it seems to me that the 450D renders more "plastique" than an M8. Don't get me wrong, sensor size is a very very important factor, but there are cases where I would take the smaller sensor over the larger. 1D3 v 1Ds, for another example.
brainiac wrote:
Now I never like to argue with eosfun, but it seems to me that the 450D renders more "plastique" than an M8. Don't get me wrong, sensor size is a very very important factor, but there are cases where I would take the smaller sensor over the larger. 1D3 v 1Ds, for another example.
Richard, we agree! I too prefer the 450D over the M8 and the 1D3 or 1DsIII decision is about more than sensor size. I just wanted to point out that this sensor size effect on image quality other than noise and so on in what I call 'sculpture',or like some people express sometimes '3D-like effect', is overlooked too often. It's also one of the reasons why some proffessionals use Phase One/ Mamiya, Hasselblad, Sinar, Cambo, etc. even though they are very well aware of the in many ways better 'pixel performance' of the D-SLR's of Canon and Nikon. They want or need that specific medium format or large format looks. Which is in those cases much more important than the high score in some DXOmark table. I believe I tasted the same scepsis to all this DXOmarking in your post. Anyway: have EOSfun
BubbaJon wrote:
No - I can see the difference in my 5D2 shots - much smoother in the shadows and the images take more twiddling than the 5D images. They are being used. What I was referring to was some converters are designed to "skip" the lower bits for speed - the last two stops (least significant bits) are a fixed value - so really not used.
Well that would be first in the DSLR market. Most of the cameras on the market use those two extra bits to digitize noise.
I'll have to re-read that article several times perhaps to really "grok" it. But my initial impressions are that he is making a few unwarranted assumptions such as the noise level being 4 bits. If you read my earlier post you'd see where I stated that (to the best of my beliefs and understanding) the minimum discernible change in level is 4 bits - not just any old 4 bits - the lowest 4 bits. His picture of the cityscape would seem to bear that out as there isn't a lot of changes apparent when you mouse over it. So are those 4 bits unimportant? No - because they are the lower 4 bits - what is important is the fact that there are 2 more bits than a 12 bit converter on the top end which expand the midtones and highlights by a factor of 2 and 4 respectively. That's the tonalities that we're really most concerned with unless you just like taking dramatically underexposed shots. In fact he more or less supports this assertion by showing how truncating the lower 2 and 4 bits posterizes an 8 bit image - those are tonal levels he's throwing away. Note that the posterization would be much less dramatic with a 14 bit image showing the full range.
I also take a bit of exception to how he describes the quantization process. Modern AD's do a pretty good job of noise rejection 4 bits are - shall we say - quibbled bits? (sorry!)
Anyway - sorry for the exposition - essentially the lower 4 bits contain significant noise only in shadows which are by definition a low signal which is significant when teh signal level is contained within the range assigned to those 4 lower bits. Whoopdy do - doesn't matter if you have an 8 bit converter those lower 4 bits still represent shadows or more accurately the lower 4 stops of your DR. More is better because it expands your midrange and upwards tonal range.
kidtexas wrote:
Well that would be first in the DSLR market. Most of the cameras on the market use those two extra bits to digitize noise.
BubbaJon wrote:
Just a quick note - 14 bits is always better than 12 bits if they're really being used. You have a better tonal range. you can always get black and white - it's those tones between you gotta worry about . It's been discussed on LL in their exhortations to "expose to the right". The general premise is accurate so I'll discuss that instead of the "gotchas". Since the range in bits is proportional to the stops of DR an interesting observation comes to bear - each bit represents a stop (as has been discussed and debated - my quibble is there's no such thing as a partial bit). So take a scene that is exposed to produce the complete range from white to black. White is represented then by a value of 2^14 or 16384. That first stop of exposure has a range of 2^14 - 2^13 = 8192 levels - fully half of your available tonal range is in the first stop (take away a bit to get the range of the next stop down). Some might say that fully half of your DR is in that first stop but that's a gray area on how you define DR - I prefer to say you have half of your tonal values in that first stop. Keep going and you find out that for each stop your tonal range is cut in half. By the time you get to that last stop you find out that it's either there or it isn't depending on how frisky the electrons felt or which side of the quantize time it fell on. So - do you consider there or not-there as a significant factor of the DR? Bear in mind we're talking blacks here and if I remember right it has been determined that 16 levels - 4 bits - is the minimum discernible between tonal values. To my mind that automatically cuts off 4 bits as a factor in DR as equivalent to number of bits. The 5D has a 12 bit converter with a theoretical DR of 12 stops. In the real world it has been determined that the actual DR of the 5D is somewhere between 7 and 8 stops depending on who runs the tests. This would bear out using 4 bits as a minimum discernible tonal range: 12-4 = 8. In the real world we use EV's to figure exposure. A rough figure is 16 for a sunny day. Roughly speaking you can speak of EV's and f-stops as the same. Anyone who has shot a sunny day scene with shadows knows that something's gotta give - typically the shadows or possibly you decide the sky can be blown. The DR of the camera is a sliding window on the exposure range of the real world - so call it 16 or 17 stops depending on your location. I'll leave it to you to draw your own conclusions on how accurate the quoted DR is for the Nikon and how well it will capture the scene.
Regards,
Jon...Show more →
Okay Jon,
So let me see if I get this correctly, a true 16 bit file (65,536 levels) can have a maximum discernable DR of 12 stops? Therefore, a 14 bit file would max out at 10 stops, and a 12 bit file would max out at 8 stops. That is, if the sensor can capture that range.
So let me see if I get this correctly, a true 16 bit file (65,536 levels) can have a maximum discernable DR of 12 stops? Therefore, a 14 bit file would max out at 10 stops, and a 12 bit file would max out at 8 stops. That is, if the sensor can capture that range.
Is that what I understand?
An oversimplification (on my part) but essentially this is what the real world has been showing - typical sensors show about 4 stops less DR than their bit depth would indicate as a *theoretical* maximum.
I was just pointed (post prior to this one) to an interesting article that I'm wading through. Part of my struggle is that he keeps using photoshop - it'd sit better with me if he was using Mathematica or something with a bit more precision. But I guess as a chalkboard Photoshop ain't bad. Anyway - his article tends to substantiate several things that have been held as conventional wisdom for some time. One of the more interesting is that at high ISO's it's better to under expose by a stop and boost in post. This has been known since the 5D came out but it's nice to see the maths and reasoning behind it. Anyway his article deals more with noise and IQ whereas my assertion is that DR is reduced from some theoretical value by about 4 stops - mostly from the perceptual fact that in shadows the DR is challenged by reduced tonal range. We simply run out of bits to adequately describe the tonal range. The world is analog and when you try to stuff it into bits there are compromises to be made.
BubbaJon wrote:
I'll have to re-read that article several times perhaps to really "grok" it. But my initial impressions are that he is making a few unwarranted assumptions such as the noise level being 4 bits. If you read my earlier post you'd see where I stated that (to the best of my beliefs and understanding) the minimum discernible change in level is 4 bits - not just any old 4 bits - the lowest 4 bits. His picture of the cityscape would seem to bear that out as there isn't a lot of changes apparent when you mouse over it. So are those 4 bits unimportant? No - because they are the lower 4 bits - what is important is the fact that there are 2 more bits than a 12 bit converter on the top end which expand the midtones and highlights by a factor of 2 and 4 respectively. That's the tonalities that we're really most concerned with unless you just like taking dramatically underexposed shots. In fact he more or less supports this assertion by showing how truncating the lower 2 and 4 bits posterizes an 8 bit image - those are tonal levels he's throwing away. Note that the posterization would be much less dramatic with a 14 bit image showing the full range.
I think its all a bit more complicated than that. He does say that quantization error/noise is pretty small. One issue with the whole 'expose to the right'/accessing higher bit numbers argument is that the noise goes up with higher exposure. Not SNR, but noise, primarily from photon statistics. Photon noise goes like the square root of the photon count. So as you go up the tonal scale, the amount of raw levels used for capture noise also goes up.
There are reasons to use the expose to the right philosophy, but not for the whole high bit/tonality reason, but for SNR reasons.
I can't recall the specifics since I don't have the time at work to reread the whole article from page one. But it all does hang together and is supported by analysis from actual cameras. The long and short of it is: "Ideally, the noise should slightly exceed the quantization step, in order that roundoff errors introduced by quantization are negligible, and that no bits are wasted in digitizing the noise." In the current crop of cameras (well, up to the 1DSIII and D3), 12 bits essentially does that.
I have a different philosophy to expose to the right:
- if contrast is within the DR range, then expose to ensure highlights aren't blown
- if contrast exceeds the DR range, then expose to show the tonal range that you are interested in as you want it to look in the final print
Somewhere between those two methods is usually right.
. It's been discussed on LL in their exhortations to "expose to the right". The general premise is accurate so I'll discuss that instead of the "gotchas". Since the range in bits is proportional to the stops of DR an interesting observation comes to bear - each bit represents a stop (as has been discussed and debated - my quibble is there's no such thing as a partial bit). So take a scene that is exposed to produce the complete range from white to black. White is represented then by a value of 2^14 or 16384. That first stop of exposure has a range of 2^14 - 2^13 = 8192 levels - fully half of your available tonal range is in the first stop (take away a bit to get the range of the next stop down). Some might say that fully half of your DR is in that first stop but that's a gray area on how you define DR - I prefer to say you have half of your tonal values in that first stop. Keep going and you find out that for each stop your tonal range is cut in half. By the time you get to that last stop you find out that it's either there or it isn't depending on how frisky the electrons felt or which side of the quantize time it fell on. So - do you consider there or not-there as a significant factor of the DR? Bear in mind we're talking blacks here and if I remember right it has been determined that 16 levels - 4 bits - is the minimum discernible between tonal values. To my mind that automatically cuts off 4 bits as a factor in DR as equivalent to number of bits. The 5D has a 12 bit converter with a theoretical DR of 12 stops. In the real world it has been determined that the actual DR of the 5D is somewhere between 7 and 8 stops depending on who runs the tests. This would bear out using 4 bits as a minimum discernible tonal range: 12-4 = 8. In the real world we use EV's to figure exposure. A rough figure is 16 for a sunny day. Roughly speaking you can speak of EV's and f-stops as the same. Anyone who has shot a sunny day scene with shadows knows that something's gotta give - typically the shadows or possibly you decide the sky can be blown. The DR of the camera is a sliding window on the exposure range of the real world - so call it 16 or 17 stops depending on your location. I'll leave it to you to draw your own conclusions on how accurate the quoted DR is for the Nikon and how well it will capture the scene.