Thanks jrista for your reply. A couple of things:
I guess I would call that making existing DR (in these cases, in the shadows) more usable, where as before downsampling it may be less usable. It can't increase DR beyond the capabilities of the hardware, however, regardless of how you work the math.
The DR the camera can capture/record is (partially) hard limited by the ADC, yes?
In other words, an 8-bit ADC would necessarily limit your DR to 8EV? Want to make sure I'm understanding this correctly. For example, if full-well capacity is 65,536e-, & you use a gain of 256e-/ADU, but now your blacks are crushed even if the read noise of your sensor is, say, 30e-. Even though your sensor may be differentiating blacks, a lot of blacks will be binned to an ADU of 1, correct?
If so, then, yes I totally see your point.
That would indeed be exactly correct.
Canon also uses a bias offset in each frame. The exact nature of it and exactly how it is used by a RAW processor is not fully known to me, but at the very least it sets a black point, allowing the total ADU range to span from -2048 to whatever maximumSaturation-2048 is. I don't know how that might affect results if you simply subtract 2048 from the original maximum saturation...to me, it sounds like you might be knocking off 2048 discrete levels of potential shadow luminance, since only the maximum saturation and stdev of noise is involved in DR calculations.
I'm not sure if a simpler strait forward equation like what LTRLI uses is indicative of real-world downsampling, either.
I wasn't talking about a downsampling equation... I don't remember LTRLI ever providing one. I was talking about his calculation of DR:
- DR = log(base 2)[max pixel value in white file/read noise in ADU]
I understand that's a general way of determining PDR, but my issue, which I raised a number of times before, is that to relate this calculation to the DR the camera is capable of recording requires a number of assumptions that I do not know are valid! I even provided a thought experiment that, if implemented into the signal processing pipeline, would make Nikon appear to have greater dynamic range even though it may not (basically: if you bin certain low signals to 0, rather than allow both positive & negative variation around a black point of, like 2048, you can reduce the stdev of pixels in a resultant black frame).
Right, sorry. I know LTRLI's formula is not a downsampling formula, what I was trying to say is I don't think it takes into account the realities
of downsampling properly. To put it another way, I think it produces a result without context, which is why it seems to produce much higher numbers for DR than would seem realistic when computing DR for a downsampled image. In its own right, taken at face value (without any context) its certainly not an invalid formula...but do the results produced by it have real-world applicability? Do they represent something real, or is the formula too simplistic? I used to believe that DxO's Print DR numbers, which are derived the same way, were simply demonstrating that once you eliminate noise, you are then realizing the sensors full potential. Kind of tough to keep believing that, though, when the D800's Print DR numbers are BETTER than the sensors full potential...its not quite like a star trek warp drive...you don't get 120% of maximum out of a 14-bit sensor.
I don't see how you can really gain anything from the act of downsampling.
All I'm saying is: take a black frame from your camera, load it up in IRIS, measure the stdev of pixels. Now downsample that file 4x or something; now measure the stdev of pixels. I just did for a black frame from my 5DIII, & here are my results:
- 100%: stdev = 5.9ADU | DR=11.4EV
- 25% (bilinear downsampling): stdev = 1.61ADU | DR=13.2EV
Wow! ~2 stops more DR! Kinda makes you question that DR formula... what say you?
Wow! So odd...
I definitely question the DR formula...although it is interesting that you got 13.2 EV. I should have actually run that calculation myself to prove earlier points...that there is some biased skew between Canon Print DR results and Nikon Print DR results. Why is it that Nikon images gain so much more than Canon sensors when doing Print DR? Especially if Nikon images already contain a near-minimum noise floor and wouldn't logically seem to be able to benefit as much from normalization, where as Canon sensors still have some pattern noise forms at low ISO, and it would logically seem that they have MORE to gain, not less, from normalization.
Well, all I can say is yeah, the DR formula is fishy, and thats the key point I debate whenever these threads pop up. But I'm just a hater for saying so, and I personally don't care all that much about DxO results, so it doesn't really matter in the end anyway. ;P
(not even Sony cameras, which use the same Sony sensors as Nikon cameras do, seem incapable of keeping pace with Nikon cameras for some reason...which is VERY ironic.)
That is interesting & to me points to the signal processing pipeline in Nikon cameras doing something different that yields better results (real world? DXO? both?... I can't comment yet b/c I haven't yet done my 5DIII vs. D800 transmission wedge test, which I'll do as soon as my friend gets his D800).
Perhaps, but the major differences only seem to exist in normalized results. The two brands are a lot closer in performance when comparing native results...which makes me suspect that there may be some kind of software-level bias involved, rather than significant hardware-level differences. There is definitely a value to producing normalized results for consistency in comparison...but DxO is a bit of a black box. They expose some details and a little bit of math about their process, but there are also a lot of unknowns. Involving software in the process can affect results in a LOT of ways that could affect the objectivity of the final outcome. Who knows, for sure, exactly what their software actually does. Does it have any brand-specific optimizations that aren't publicized? If so, how are they skewing results?
The ironic thing here is that you can completely ignore the normalized results, and the native hardware results still indicate that Nikon and Sony sensors are technologically superior. Thats all someone really needs to say, and its not something that can really be (nor should be) disputed...but the Print DR results and funky S___ like "bonus points" are involved in the final "scores" that DxO gives to each camera. I find that to be fishy as well, and unnecessary. You don't need biased results to demonstrate that Sony Exmor based cameras offer superior IQ, and the actual meaningful differences between the best of the vest (D800) and the best of the competition (5D III or any one of the measured Digital MF cameras) are not really as big as DxO is making them out to be. Thats where my pet peeve lies...bias and skew when its unnecessary and not really honest or helpful to potential customers of any brand.