Crop Factor, Low Light and Aperture with m4/3 lenses? Part 2

Started Jun 16, 2014 | Discussions thread
knickerhawk Veteran Member • Posts: 5,952
Re: Reply to Just another Canon shooter

Just another Canon shooter wrote:

knickerhawk wrote:

Just another Canon shooter wrote:

knickerhawk wrote:

DXO 6D: 45.32

DXO Oly: 43.22

LR 6D: 40.55

LR Oly: 42.93

Lower Std. Dev. in LR for both cameras and the differences between DXO and LR is GREATER for the 6D than it is for the Oly. Hmmmmm...

It should be, the crops are not normalized to the same area. To normalize, you need to shrink the 6D ones about 14.5%. Then they will look smaller (because dpreview frames for the same height, not the same diagonal). I will leave the calculations for you, and you may very well discover that the 6D wins.

The fact that you think I need to normalize the results shows that you're missing the point. You claimed that Adobe is applying (greater) NR to the Oly but not the 6D. If that were the case you would expect to see a larger difference in std. dev. between the two Oly samples than you get for the two 6D samples. The opposite is the case.

St.Dev. is a measure of how the luminance values are distributed with disregard of their location.

Nope, it's not just luminance values.  I gave you the composite values before, but you can also measure the std. dev. for individual color channels in RGB or the individual L,A,B channels.

The nature of the noise (the "pattern") is a property of the spatial distribution, which is a completely different property. I was talking exactly about that.

While spatial distribution is a factor to consider when examining noise, it's not the only consideration. The std. dev. measurement won't tell you anything about the spatial distribution but it will tell you something useful about the amplitude (not sure if that's the best term to use here), which is just as important.  Bottom line: samples of the "solid" colors in the color checker portion of the test shots have increasing std. dev. measurements for increasing ISOs, which is predictable and expected.  The std. dev. doesn't tell you everything about the character of the noise you're seeing in what's otherwise supposed to be a uniform patch, but it's still useful objective information for comparative purposes.

Does the crop on the lower right look NR'd to you or no? If it does, this is an indirect confirmation that the St.Dev. is not a good metric of what you see.

If you actually look carefully at your respective patches, you'll see that what differentiates the noise present in the LR Oly patch relative to the DXO Oly patch is a bit less luminance noise but also a bit more color variability.  The std. dev. measurements for the LAB channels confirm this by showing a lower std. dev for the L channel in the LR version relative to the DXO version but higher std. dev. in the two color channels.  Thus, if you're going to say that LR is applying (luminance) noise reduction, then you must turn around and also say that DXO is applying (chroma) noise reduction for the Oly.  Or you could just stop trying to needlessly complicate the discussion with these groundless cliams about Adobe adding "noise reduction" to its Oly raw conversions...

By the way, with respect to the 6D, the LR conversion also does better overall and with respect to the L and A channels.  DXO did better with the B channel only.  Again, if there's any basis for claiming "noise reduction" in the Adobe demosaicing algorithm, it would be more applicable to the Canon than the Oly.

To sum up, the std. dev. measurements and my own two eyes confirm that the character of the noise (the look) is different between these two cameras and the two converters but that's hardly a surprise and certainly not evidence that Adobe is somehow inappropriately tipping the scales in favor of the Oly (and only the Oly) with respect to noise reduction.  Your speculation on that point is meritless.

Did you follow the discussion so far? This is the noisiest color for the 6D.

Yes, I've followed with considerable interest the "evolution" of your argument in this thread. I'll be interested to see where it evolves to next.

You'd better go to the past, to part I, if you are really interested.

As I said, I've followed the evolution of your argument from the beginning of the original thread.  By the way, according to DXO, the red channel is the least sensitive for both the 6D and the EM1.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow