I understand normalization perfectly.
It seem like you do.
I do not think it is valid in all contexts. Noise frequencies are one thing...but that does nothing to tell you about editing latitide in a RAW editor like Lightroom. YOU CAN NOT EDIT DOWNSAMPLED RAW. That's a misnomer. Downsampled RAW images do not exist.
What does this have to do with fairly judging whether one camera is better, worse, same in regards to DR,SNR??
Your talking noise frequencies. I'm talking editing latitude.
Yeah because you need to take into the account the former if you are trying to make statements like one camera has better SNR or DR than another and what editing latitude you have when trying to take max advantage of MP account is something else entirely.
The problem here is not that I do not understand normalization. It's that you refuse to look at the problem of comparing cameras from a different angle than the one DXO has imposed upon you. ;P
No, it's that you don't understand why normalization must be carried out to make statements about what camera does better than another for certain things.
Your misunderstanding. DXO's results are only meaningful when you are on DXO's site (i.e. isolated) comparing cameras that DXO has tested. Many of DXO's results and measurements have no relevance outside of their site, in the real world...such as, oh, say, lifting shadows in Lightroom. Lifting the shadows of a RAW image, an UNSCALED RAW image, in Lightroom?
Relatively they do!
What if I want to know how to cameras compare IN THAT SPECIFIC CONTEXT? Well, Print DR is invalid, it doesn't have the capacity to answer my question in that context. Screen DR, on the other hand, DOES.
100% misleading, who cares if some 1000MP camera can't pull the shadows well when images are viewed, in all 1000MP glory, at 100% view, compared to some 10MP camera viewed at 100% if if you viewed the image from the 1000MP camera at the same scale as you view the 10MP image the 1000MP camera gave you better SNR and DR???
It tells me the dynamic range of the full sized, unscaled RAW images. I WANT to COMPARE that between cameras. That is NOT an invalid goal. On the contrary, THAT IS WHAT EVERYONE CARES ABOUT WHEN THEY THINK ABOUT DR!!
That isn't fair to lower MP cameras. As I say, you may be able to maintain the same detail as a lower MP camera and have better SNR and DR even though viewed at a scale where you take advantage of all the extra detail it might, at 100% view, measure worse.
Yeah maybe you want to know how you'd do, taking full advantage of the new camera's resolution compared to what you were getting when you took full advantage of your old, lower MP camera and get a sense of that fine, but at the end of the day it's all the same not fair to slag off on the higher MP camera and say it is worse or not as fully better as it is, normalized to the same scale as your old camera. You might end up thinking that the new model flat out would give worse SNR and DR and perform worse than the old one when it really might not at all or you might minimize the amount that it is better and so on.
Do you get it now?
Now and before.
When it comes to shadow lifting, the number of megapixels doesn't matter. The dynamic range of each and every pixel is what matters. I don't really care about the photon shot noise levels, which permeate the entire signal. I care about the READ NOISE levels, which only exist in the deep shadows. In that context, it is entirely fair to compare across cameras, because what I want to compare is only valid at full resolution. The frequencies of all noise are immaterial, the RMS level of the READ NOISE is what matters.
so read shadow noise is magically invariant on scale???
You can only compare cameras using information produced the same way. Print DR, on DXOs site, is only valid when comparing cameras within the context of DXO. It is entirely invalid to use the Print DR value from DXO, and compare it to any dynamic range value derived anywhere else, say DPR. Print DR only gives you a numeric value with which to compare cameras in one specific context...DXO. It does not give you any real-world information beyond that context.
oh brother, yeah the exact numbers taken alone aren't generally useful but comparing the numbers relatively they are, the relative differences are absolutely generally useful in outside contexts that is the entire point of normalization!
I am not trying to trick anyone. I believe DXO IS tricking people with their Print DR numbers...they are regurgitated all over the net, OUT OF CONTEX, ALL THE TIME...and that is exceptionally missleading.
Maybe you are not trying to trick people, but you are misleading them by mistake then.
And yeah notice how basically everyone else does use those numbers? But nope, Jrista is the only person the world who actually 'understands' and 'gets' normalization. I mean, OK, here and there a few people make the mistake and assume those numbers are what you'd get just dealing with a RAW file at 100% alone, and it's true for that, to get the true number, in isolation, Screen DR is the one to go, and a few people mix it up and use Print DR for that too, but not that many from what I see.
True, for most things. Not true for shadow lifting.
Yes, because shadows are magical unicorns.
It is fair! I can't lift shadows with a RAW image that's been downsampled...because I can't downsample a RAW image. I have to convert it to RGB pixels, then downsample it, then save it as, say, a TIFF. The TIFF doesn't have even remotely close to the same editing latitude.
Who is talking about having to edit the TIFF afterwards???
I could care less about the rules DXO enforces on "comparing" cameras. I know what normalization is, and they provide useful details in some contexts. But that is not the context I am usually referring to. There is more to comparing a camera than ONLY comparing JUST the sensor, and JUST a normalized output at that. There are far more things and ways to compare than just the normalized image context. I'm not saying comparing in a normalized context is invalid...it's just incomplete.
Of course there is more to comparing than just the sensor, but we happen to be talking about just the sensor here.
Outside of that context, a dynamic range value of 14.4 stops of DR for the D800 is invalid. When people get into lengthy, extended debates about the shadow lifting range of the D800, they should be using 13.2 stops of DR as the reference...which would mean they could lift a bit over 5 stops without seeing noise in the shadows. That is exactly what a lot of the examples on DPR indicate...but the debate still rages on, why? Because DXO says 14.4 stops.
If they are viewing full detail at 100% then yes they should only expect to pull 13.2 stops. But if they want to know how many stops better it might do than say a 5D2 it would be the 3 stops relative difference not just the 2 (or whatever the exact numbers are, you get the idea).
Anyway, were talking at perpendicular angles here. I understand normalization. Normalization has it's place. Normalization has it's use. When it comes to discussions of dynamic range an the shadow lifting ability of cameras, Print DR is invalid. Screen DR is valid. If you want to COMPARE the shadow lifting ability of cameras, then Screen DR is the value you have to use.
No it's when you are relatively comparing that you must use the PrintDR. It's the ScreenDR that you use when you just want to know how much you can pull and lift when viewing the RAWs at 100% view.
I concede the point about normalization for comparing "fairly" as you say. I've never denied it.
You jsut spent the last 10 paragraphs denying it.
But that is different than what I'm talking about, and it ignores the constant debate about WHAT DYNAMIC RANGE ALLOWS in cameras that have more of it (or, to be more precise, allows in cameras that have LESS READ NOISE...because that is primarily what were talking about here...the difference between the D800 and 5D III isn't sensor (pre-read) dynamic range...it's read noise levels (post-read).)
Yeah keep trying to conflate what you can get out of a RAW image when working on it at 100% view and what to expect in that context with comparing how sensors do relative to one another. For the former, yeah it is the ScreenDR, for the latter though it is PrintDR.