Comparing FF vs. APS-C

[Jayson]

Great tests John. I bet you put a ton of time into that and I thank you.


Jayson

[Kayaker72]

Like the others, I very much appreciate the time and effort you

[Jon Ruyle]

it also seems to be decreasing the sharpness of the 7D images significantly more than the 5DII. Anyone have an idea as to why?

That is no surprise: the 7D pixels are much smaller, so flaws appear magnified.

[HDNitehawk]

it also seems to be decreasing the sharpness of the 7D images significantly more than the 5DII. Anyone have an idea as to why?

That is no surprise: the 7D pixels are much smaller, so flaws appear magnified.

This may be true, but I will givemy theoryas well:


The 70-200mm will produce different results at diffrentpoints. I find the 200mm end to be sharper than the 70mm end. So in the comparison that John made it was a good comparison for noise. But in this test I think it would have been biased for sharpness moretowards the 5D II more so because of the different lengths used. If you ran other scenarios where you used the same lens at different distances, I think the test would be biased to the 7D because of its greater DOF it would appear to be sharper. (whether it actually would be or not is a different story)

[Jon Ruyle]

I find the 200mm end to be sharper than the 70mm end.

That, too. Definitely.

[neuroanatomist]

Thanks for the feedback! FWIW, the approximate focal lengths on the zoom were 85mm for the 7D and 135mm for the 5DII.

[neuroanatomist]

I added some larger composites (1600 pixels wide) toFlick<span style="color:#ff00ff;"]r photo tests page. A new composite looks at color - I was surprised to see the amount of desaturation in the red channel with increasing ISO (click image for full size).

[genotypewriter]

If you're doing a colour test, there are far too many variables here and also I'd question some of the choices (e.g. tungsten lamps and AC ones at that shot at 1/60... unless that's only what you want to know about).


For example you should have used the exact same area of the image circle of the lens at one focal length when doing the test. Although we don't always notice it, colours change at different points on the image circle, focal lengths, apertures, etc. Since this is not a resolution/noise test, you can always keep both cameras at the same distance, same focal length+f-number on the zoom and then compare the colours.


And the use of a preset WB is questionable, because it's likely to be different between the cameras. If you really want to know how the two sensors behave under such light, you should have no WB adjustments applied.


Also I'm not sure whether the ND filter use is a good idea (stacking aside). Sensors have different response curves and the transmission curves of the filters can make things favourable for one camera and bad for the other. So with a different brand of ND filters, you most likely will see a different outcome. Also the filters will add their own flare (reducing the contrast) and color casts (hopefully evenly, but unlikely... another reason to use the same area of the image circle) but at least it will be constant for both setups.


Constructing the test the way I'm suggesting could be perceived as being "too artificial" to be relevant for every day use but, then again, so is this. I see highly controlled test setups as a necessary evil to understand the most fundamental/constant differences, whether or not they will be obvious in general usage.


Good luck!


GTW

[HDNitehawk]

Interesting John.


I have noticed that in certain situations the 5D II can have a problem oversaturating in red. For most types of pictures red isn

[neuroanatomist]

Thanks for the feedback, GTW. It wasn't intended as a color test, but the desaturation just 'jumped out' at me as I looked at the images. The desaturation with increasing ISO does not seem to be different between the two bodies, so the concerns about using the same area of the image circle and different preset WB are obviated when just looking at increasing ISO on one body.


Incidentally, though, you're quite correct about the WB being different between the cameras - although DPP does not show the actual color temperature of the WB settings, DxO does. For tungsten WB, the 5DII uses 3455 K / +4 tint, whereas the 7D tungsten WB is 2527 K / -21 tint. Likewise, the fluorescent WB settings are 4292 K / +18 for the 5DII and 3990 / -14 for the 7D. Still, I think a preset might be the best option - the presets are set such that a light source of a given temperature (e.g. 3200 K for tungsten, and presumably Canon uses the same source temperature for all their cameras) render light from neutral gray object as neutral gray. That should compensate for differences between the sensors (Bayer masks, actually) in terms of their color sensitivities. As a practical note, how can one, "have no WB adjustments applied?" There will always be a color temperature associated with the image - even though it's a metadata setting, it's always used to display the image. Presumably, you mean setting the color temperature to the same value (e.g. 3200 K for tungsten) for both cameras - but in that case, neutral wouldn't be neutral anymore (which is the whole reason tools like the ColorChecker are available, right?).


I did give some thought to the use of ND filters. The reason for that was to keep exposure (aperture and shutter speed) constant across all the ISO settings, since I was primarily looking at noise. I agree that with a different brand of filters, the results would be different. With the B+W filters, at least, they publish the transmission curves - for the 3 filters that I used #101, #102, and #103), they are reasonably flat across the visible spectrum. If anything, transmission in the red/far red part of the spectrum is slightly higher (an effect that gets more pronounced with the higher density filters) - so in theory, the addition of the ND filters should slightly enhance saturation in the red channel (i.e. the desaturation might be worse than seen in this test). Still, it's a variable I'd like to control for, so I will probably reshoot just the ColorChecker with one camera, varying the shutter speed with ISO instead of using filters to control exposure.

I see highly controlled test setups as a necessary evil to understand the most fundamental/constant differences, whether or not they will be obvious in general usage.

I agree - it's important to keep the perspective that contrived tests only show a small part of the truth. I like Bob Atkins' analogy of representing the quality of a lens with a single number, e.g. an MTF50 averaged across the frame, which he likens todescribing the Mona Lisa by it's average color. Still, in this case, cranking up the ISO of an image to boost sensitivity (which is in keeping with the ETTR philosophy) might have the unintented consequence of negatively impacting the color palatte of the image - and that's a real-world consequence.