sandpiper wrote in post #16110668
So, you are saying that a shot taken at ISO 200, three stops underexposed will yield as good an image (after editing) as one taken at ISO 1600 and exposed correctly??
At ISO 200? On a Canon sensor? No. That's the nonlinear region.
But an ISO 800 shot underexposed by 3 stops might, since that's in the linear region. Honestly, I'd have to try it out.
Thing is, though, that ETTR is based on the notion of the higher ISO sensitivities working better than pushing the lower ones.
There's also the signal to noise ratio to think about. At ISO 1600, the 40D is already at the point where the falloff is at about 3 dB per stop. The 7D's falloff is about 2.5 dB per stop at the same place. But at ISO 3200, the 7D's falloff rate is 3.5 dB per stop (all of these are from the "in print" measurements). One stop of difference is equivalent to 3 dB, so that suggests that you'll actually get better results on the 7D by taking an ISO 3200 shot and pushing it rather than using ISO 6400 and above.
Each camera, therefore, will have a different point at which you'll get equal or better results by pushing a lower ISO shot than by taking a higher ISO shot.
So based on this, the experiment you mention, even at ISO 800, will work out in favor of the ISO 800 shot, though not necessarily by a whole lot. That has nothing to do with the nature of going to higher ISOs, however, and everything to do with the characteristics of these sensors and their readout electronics. Once you get the camera to a certain ISO, going beyond it in-camera appears to yield no real returns except for ease of use.
Of course, all this also assumes that dxomark.com's sample cameras are representative. The only way to know how your specific camera will actually behave is to perform some of these experiments yourself.
How do you deal with clipped blacks that have gone off the edge of the histogram due to the heavy underexposure ?
The dynamic range prevents the blacks from truly going off the edge like that. You have to be shooting raw to get away with this, though.
How do you think the 30D and 40D are able to do the equivalent of ISO 12800 at all?
The higher your ISO, the lower your dynamic range. This is something that dxomark.com actually measures. You can see the dynamic range graph for these cameras for yourself. Take a look, especially, at the "print" graph, where the cameras are equalized (and thus comparable on an even footing). Note how the dynamic range falls off with ISO.