Canon EOS 5D MkII, Nikon D3X, and Sony Alpha 900 compared
This set of full size shots was taken with the still life left set up, because the Nikon and Canon cameras were not here at the same time. The report originally appeared in December 2008 on photoclubalpha. It compares the A900, 5DMkII and D3X using the converters supplied by the makers – Image Data Converter SR2, Digital Photo Professional, and Capture NX2. Each small image in the article can be clicked to open a Level 10 quality full size JPEG – beware, the largest is over 13MB of data.
All images are from uncompressed raw files at the maximum bit depth selectable (if a choice exists), at ISO 100, using ‘Standard’ picture look default as defined by each maker. NR and all other similar controls are turned off, and sharpening is set to the minimum or disabled in every case.
Alpha 900 with 100mm Minolta macro (click image for full size file). Minus -0.35 stops exposure correction needed in IDC2 raw conversion to fairly match the exposure set using the Canon, below.
Canon EOS 5D MkII with 100mm Canon EF macro (click image for full size). Uncorrected raw exposure in raw conversion via DPP.
This image was taken a week after the others; the lighting and position have been replicated (the lights were left undisturbed but the viola was played, so had to be positioned by matching a print from the Canon. The Nikon DX3 was fitted with a 105mm f/2.8 Nikon AF-S VR Macro lens, at f/11; the ISO 80 (Lo-0.3) setting was used with Capture NX2 shadow and highlight adjustments for a fair match to the in-camera JPEG and the other results. Click image for full size.
The point of focus, using the centre focus sensor for each camera, was the end of the fingerboard. Actual focus showed variations between repeated shots with all three cameras, tripod-mounted and working with two Elinchrom BXRi 500 flash heads. The exposure was set for the Canon and adjusted to be perfect at f/11. However, the Sony JPEGs indicated slight overexposure and the Nikon ones even more; -0.35 exposure compensation was used when processing the Sony files, and -0.65 for the Nikon. For comparison, a Nikon shot taken at the ISO 50 setting, with no exposure adjustment, is included.
Alpha 900 with 28-105mm RS Minolta lens at 105mm (identical tripod position). Throughout this test, I noticed that the Sony could almost have been used with a polarizer, relative to the Canon. For some reason, the Sony suppresses specular reflections and reveals more colour in the wood and fabric. It is not just an exposure and contrast difference (and of course, it was not used with a pol). All cameras were set to Daylight WB, but their definitions of this differ. The Sony and Nikon both have a relatively warm colour, the Canon is very neutral.
Canon EOS 5D MkII with 24-105mm IS L lens at 105mm. Again, not as tight as a true 100mm macro – and both the macros do use internal focusing, so it’s not a matter of real lens extension versus internal focus fakery. Or not entirely to do with that!
Nikon D3X with 24-120mm VR Nikkor at 102mm (composed by eye, tripod moved to allow for thickness of camera grip-base design). Uncorrected conversion (required around -0.65EV to match the Canon brightness) in Capture NX2. See ISO 50 version which follows.
Nikon DX3 uncorrected ISO 50 exposure (a little darker than the Canon ‘correct’ original setup). Nikon does not claim an exact EI 50 for this setting, it is just marked as -1 on the ISO range. The flash is repeatable to 1/10th stop, and with experience of using other lenses on the cameras, the general confirmation would be that the Canon is around 1/3rd stop ‘denser’ in typical raw and JPEG results than the Sony, and the Nikon about 1/3rd stop ‘brighter’. Since the Sony has a very much flatter tonal curve rolling off the highlights (preserving highlight detail) this can not be converted to ‘ISO ratings’, but has practical implications, and each camera should be tested for calibration with a studio flashmeter for use in this environment.
The lens choices
The Sony used a Minolta 100mm f2.8 AF (1986) and a 28-105mm RS f/3.5-4.5 (1999). At 105mm, this lens still did not crop as tight as the macro at 100mm. The Canon used a Canon 100mm f2.8 EF macro and a 24-105mm f/4 IS L. Again, even at 105mm, this lens was not matched at all to the 100mm view. The Nikon, with no macro available, was tested using just the 24-120mm VR f/3.5-5.6. This was matched visually to the 100mm crop, and the focal length reported was 102mm. My lens choices were dictated partly by what was available for loan review from Canon and Nikon, and what lenses I have decided to keep for my Alpha 900. The 28-105mm RS Minolta lens may come as a surprise, but has replaced the 24-105mm D Minolta in my 900 outfit; it’s just a much better lens all round desite being about a decade older.
Focus and depth
f/11 is inadequate, with files of this size, to secure enough depth of field for 100 per cent viewing. But f/16 and f/22 (etc) reduced sharpness greatly by diffraction, and some tests at f/8 gave much better sharpness – with so little depth of field they would be unacceptable. So f/11 was selected as the best compromise, showing d-o-f, bokeh, minor focus errors, and being an aperture at which all the lenses used should be up to the demands of the sensors. All three cameras offer lens specific focus adjustment; it did not prove possible to improve consistency on this subject by using the micro adjustments. The A900 was consistently the most accurate in focusing as targeted. All three cameras had the centre AF sensor only active, in single shot mode.
This test shot highlights the need for tilt lenses (or tilt-shift) in the studio with high pixel count full frame DSLRs. Ideally, the shot would be taken at f/5.6 or f/8 with a high quality 90mm Tilt-Shift tilted to align the plane of focus perfectly so every part of the viola from tailpiece to headstock was sharp. Nikon and Canon both offer exactly such a lens; Sony does not. I have a tilt adaptor (ARAX, from Kiev) and two lenses – 50mm Pentacon and 80mm ARSAT – to fit this. These are no substitute for a good range of Zeiss 24mm, 45mm and 90mm TS lenses to fit Sony Alpha!
In practice, you can stop down to f/22 if you want. Though diffraction effects reduce sharpness, it can be restored by careful use of raw conversion capture-stage sharpening to a level which betters any scan from rollfilm. Diffraction sharpness loss is not the death of a decent digital image, it is just another factor like differential focus blur; you can trade it off against depth of field. Do not limit your technique armoury by refusing to use apertures smaller than the supposed diffraction limit!
The image above is one of my first shots taken with the Alpha 900 – at f/22 on the 17-35mm Minolta D lens. Some strong detail recovery sharpening has been used in ACR when processing the raw file. It’s not perfect but proved to me I could rely on my existing 17-35mm and use the same gamut of depth of field techniques I once used on film.
On an iMac 24″ 2.16GHz running OSX 10.5.5, the Sony IDC2 software was fastest for viewing and most stable (used in conjunction with its companion lightbox app). I had not rated it highly beforehand, and was surprised by the improvement over older versions. Nikon was stable and very fast to save files, but slow to view and to build 100% views for checking focus. Canon’s DPP proved unstable, exit-crashing when asked to handle more than 1000 previews (517 files, raw+JPEG) but OK when prefs were changed to show raws only.
All three programs were relatively tedious and slow in use compared to Lightroom or ACR. Canon appeared to apply a more subtle sharpening and a very steep midtone curve. Sony’s sharpening was coarse and when set to its minimum (not zero) did not seem to aid fine detail recovery (checking these files in ACR revealed better detail, but the Canon files were also improved by ACR). Nikon’s files could not be checked yet using ACR, not supported at the time of this test.
Despite the speed of IDC2, I felt it did the images no favours. Minus 100 sharpening does not prevent fairly ugly coarse sharpening appearing in the shot, and I am told that IDC2 recognises whatever in-camera sharpness was set, and uses this as the very minimum. Since part of this test involved using the defaults of the cameras (Standard picture setting, and no adjustments to contrast, saturation, sharpness etc) plus zero optional processing (NR, DRO, D-Lighting, Lens Shading compensation etc) IDC2 has slightly skewed the view of the raw files.
Why this test?
I had been using the Canon 5D MkII for magazine reviews for two weeks, and the Nikon D3X arrived on the day the 5D departed (again, for a magazine test report). It was necessary to set up something for a planned article comparing the three cameras. Like most of my tests, it is practically based, not a test-chart exercise. I set up and light a subject as if I was taking it for real, and use the camera as I would expect to. I then see what I get.
All I can tell you for sure is that any one of these cameras will do this job perfectly and the differences you see are insignificant in an A2 reproduction, whether from my Epson 3800 (seen) or in litho print (unlikely to happen!).
– David Kilpatrick