So I was always under the impression that RAW files are very similar to one another, despite camera model or manufacturer, and all the differences were due to software interpretation of the raw data.
A few days ago I decided to play with some RAWs from DPReview’s camera comparison studio scene. I was shocked to see how different they were. I have attached a screenshot with the pipeline I used (for exposure and color calibration I used the color checker, for AgX auto tune levels). Also here is a comparison between D610 (my current camera) and α7 III.
There seems to be a lot more contrast in the one image and just wondering likely you should match the exposure so maybe the black and white patches to see where you end up…
It can seem this way when using software like Lightroom as they tend to use camera-specific DCP profiles to give a more consistent starting point. (To the best of my knowledge)
But in actual fact, as already stated each sensor, gain circuits, AD converters and firmware can give pretty massive differences.
Of course each manufacturer’s proprietary development does indeed give big differences too!
Also in cameras you have picture styles such as portrait, landscape, neutral, vivid etc which changes the profile used by manufacturers to produce out of camera JPGs so this is another source of difference between manufacturers and how people interpret the raw file should look. Lightroom tries to imitate the camera manufacturer’s look and many people like this. On the other hand DT gives you the freedom to create your own look.
To make my editing easier I have started to make styles for each of my cameras that give my raw files a starting point similar but not identical to out of camera jpgs. They are sharper and have more detail than the camera’s jpg but have similar contrast and color tones.
Well, they are not that different with you workflow.
I’ve seen way worse in other softwares who relies on DCP or ICC profiles and are all over the place, believe me (Capture One, for instance). I mostly see some luminosity differences between the shots, the lighting doesn’t seem equal at all between the 2 shots, the D610 is more illuminated in the borders / corners so it definitely plays a role as well here.
On the colorchecker, the red patch is more magenta on the D610 shot, and more orange on the A7III shot. All the other patches show some slight variations, but you can’t make them totally equal.
Forget the portraits btw, they fade over time, the Dpr team reminds it regularly.
Anders Torger, the man behind the calibration software Lumariver, says in his manual that the profile is roughly responsible for 90% of the rendition, the camera is responsible for 10%. If you manage to homogenize the light on the colorchecker between the 2 shots, what will remain will be the 10%.
Raw files differ in the color filter pigments in front of the sensor. This is mostly compensated by the initial color space matrix, which is supposed to translate the raw sensor readings into colors. But it’s not perfect and some differences remain.
I thought about this too, but it seems to be happening with the same lens design (for example with 85mm f/1.4 G on D850 vs D610) so lens sample variation having such big differences between all the major manufacturers seems a bit strange.
Thank you, I didn’t know that about the portrait pictures. I don’t know if it is the pictures fading or different lighting though, because D610 was released in 2013 and α7 iii in 2018, so I would assume the α7 shot came later. I should look more into their process.
Again, I calibrated the exposure and white balance using the color chart. Is there something I’m missing in this workflow, or it’s just the 10% difference? Because I really don’t want to buy a Sony .
use the profile correction for the lense, especially the vignetting, and hope that it’s correct, before comparing the colors
manually tune the white and black levels in ag-x until the white patch and the black patchs show the same luminance, don’t rely only on the auto levels
if you really want to equalize even more, modify the shape of the curve (using toe and shoulder) until all the neutral patches show the same luminosity between the 2 shots, but good luck with that (I didn’t try, it’s a rabbit hole)
use the calibrate with a color chart inside the color calibration module : by default, the one in the D610 is considered better by the tool (less delta E). Both profiles are slightly better after calibration, considered as “good” (A7III) and “very good” (D610). Which means that the differences are objectively not that huge.
The chart is in a border and that is not the correct way to calibrate a sensor, it should be in the center to avoid the effect of vignetting. That’s why in this particular case, I would advice to correct the lens before calibrating the colors. Because the lenses are different, the vignetting values are probably different as well, and it will impact the lightness of the patches unless you can perfectly correct the vignetting.
The D610 uses a Sony sensor, I saw in another thread a link by Glenn Butcher to some sensitivity data of different cameras : Open Film Tools SSFs
There is no D610 and A7III in the list, but the D750 and the A7r are there, and you can see that the CFA response seems very close between the 2 cameras. If we extrapolate, it probably means that the CFA used for the Nikon cameras in this era were not that different from the CFA that Sony used.
I did a quick test using some commercial softwares : using DXO PL8 to correct for vignetting, because their lens profiles are often praised, export in linear DNG and use lumariver after that to create 2 ICC profiles for Capture One. They are very close… the D610 is still slightly better. But the slight delta E in hue are always in the same direction for both sensor. See the screenshot :
Once calibrated, I would probably safely use a A7III alongside a D610 without worrying to much… The difference would probably be smaller with a proper target shot, in the center of the frame, with the same exposure. The colors of the colorchecker can fade over time as well, I guess…
And another thing: use as few modules as you can, to limit the number of variables.
E.g. in this case, I’m not sure you need AgX or highlight reconstruction:
most of the image is diffuse reflective with one illuminant. The areas that could be too bright are the metal objects and the highlights on the glass. But those are irrelevant for the color rendition.
The most striking difference I see is simply exposure, one image is darker than the other. AgX could interact with that (shift in the highlights), so the first thing I would do is remove the tone mapper. It will lead to bland, but comparable photos.
That should not affect the raw files though.
Dpreview claims that they are using “good” prime lenses for these images, but I agree, the lenses should matter for the fine detail, especially on the edges. Simple field curvature would affect these tests a lot.
I think it’s evident there was movement in the scene, and probably the lights have been repositioned, as well. Check here where the shadows are in the painting bowl (or whatever it’s called).
Such changes would affect all reflections, including glare.
Lens corrections don’t seem to have been applied (the distortions are different), and if vignetting is not corrected, that would also contribute to to apparent brightness, especially near the edges and in the corners.
BTW, I’m not really sure if it’s OK to post those images here:
Please do not reproduce any of these images on a website or any newsletter/magazine without prior permission (see our copyright page). We make the originals available for private users to download to their own machines for personal examination or printing (in conjunction with this review); we do so in good faith, so please don’t abuse it.
I’ve removed the test shots from the original post since we can’t host them. But this post has some interesting things so I don’t want to completely remove the whole thing.
So I applied lens corrections and adjusted AgX white/black point using the color chart, and now everything looks preety close.
I also agree that some changes in lighting should have taken place. Also noticed that all the test shots with contrasty portraits are from 2017-2018, while all the previous and later ones appear faded.
Thanks to everyone who contributed in this thread.
.