FujiShooterCY wrote:
saltydogstudios wrote:
One thing I am not sure about is where exactly this adaptation is done in the software. Is it done with some numeric coefficients in the demosaic algorithm? I mean - given two different Bayer sensors, which are both being demosaiced with the same algorithm, is adaptation done at the level of demosaic or at the level of input profile, or both?
Adobe has a white paper that sheds a bit of light on this (no pun intended). What they call "colorimetric interpretation" is likely the step we're interested in.
https://www.adobe.com/digitalimag/pdfs/understanding_digitalrawcapture.pdf
Ok, but I feel they are talking about the step when the sensor input profile is applied, right after WB, aren't they?
Gamma correction is about tone curve, am I correct? ...and tone curve is sensor-agnostic already, though each software either has a sensor-specific tone curve default, or some "generic" tone curve for the sensors it isn't already "aware of". Correct?
...the same goes for NR - you can apply a profiled NR if you have a corresponding sensor noise profile at your disposal already, otherwise you are left with "generic" NR algorithms to play with.
So while it's accurate to say that the OP is comparing RAW Converters. The OP is also necessarily also comparing sensors, because the RAW Converter must first measure the sensor - if it is to have any hope of producing a consistent result between cameras. (RAW values to RGB values.)
But Ok we are comparing the grade of software adaptation to the sensor again, not the sensor itself.
I suspect that camera manufacturers tune their sensors for different goals.
Hmmmm do you mean "tuning" at the CFA design stage, where color filters spectral curves are designed?
But Ok, even if this is true, doesn't sensor input profile level the difference?
I mean, let's conduct a thought experiment. Imagine we take 4 camera bodies, i.e. Nikon, Canon, Sony and Fuji (btw three of the four are actually using SONY-manufactured sensors). Also we take one single lens sample of a model which is adapted to any of the four with a dummy adapter.
Now we take an IT8.7 target, take RAW shots of it with each of the bodies under the same (proper) conditions, and generate sensor input profiles - one for each body, so for each of the bodies we get the colors and tones of our target reproduced 1:1.
Now when we take 4 photos of the same scene with the same lens using our 4 camera bodies, and push the obtained RAWs into the developer software, what we will achieve?
All four will overcome the demosaic processing first (when profiling, we did this too).
Then RAW exposure correction, maybe with the interaction of the software default settings (when profiling, we did this too).
Then our input profiles (carefully crafted earlier) are applied.
Then we take care of the tone curve and ensure that all 4 shots are treated with the identical tone curve, but not the one which is nailed to the specific sensor in the software default settings.
And, finally, we are displaying all 4 images on a single screen (monitor) side by side.
My bet is, we will get the resulting images almost identical and indistinguishable by the naked eye without the precise instrumental measurement.
But you are saying that, quote:
To my eyes Nikon cameras produce more filmic skin tones in natural light - less color variation around the central skin tone. Canon cameras produce more color variation around the central skin tone. Leica cameras produce the most (though I have no hands-on experience with Leica cameras).
Can you please explain, how can this happen under the conditions listed above, given input profiles that are leveling the differences in sensor spectral sensitivities, and the identical tone curve?
I prefer Nikon for natural light and Canon for studio work. The Fuji X-Trans1 sensor is the best all-arounder (see below). Leica is trash at skin tones (yes I said it).
I read somewhere that Canon produces more accurate sky colors than Nikon.
My guess is that you are actually talking about the differences in a) sensor input profiles, b) default tone curves that are nailed as defaults in different softwares (in-camera and/or external).
I'm aware that there are software packages that will do RAW conversion without having a deep database of sensors - I have a few installed on my computer.
Very interesting! Can you please share your experience, which convertors are sensor-aware and which aren't? Thanks!
Adobe software refuses to convert any camera they haven't measured,
What do you mean by "measured"? Am I correct in my assumption that you are talking about camera-specific sensor input profiles available (or not)?
which leads to the odd DP Review "initial hands on preview" statement "We can't evaluate the RAW files because Adobe hasn't released an update for it yet." (which I always thought was odd since it implies that Adobe is the only way to evaluate RAW files...)
Yup really, the DP Review approach to "testing" is somewhat questionable at least.
I.e. how do they compare camera bodies for sharpness/resolution, given they use different lenses with each body? The correct approach would be to take one single very good sample of one single lens, and adapt this single sample to every camera body under the test, yes?
According to RAW Therapee they support a wide variety of cameras, but if they haven't measured the camera the colors may look off.
Again, if they aren't talking about camera-specific sensor input profiles, then what to they mean by "measurement of the camera" than? Can you please shed some light at it?
Though because it will open RAW files for cameras that are newer than the software, I suspect they have a generic fallback color profile for the camera manufacturer.
Yes, definitely they do have one.
https://rawpedia.rawtherapee.com/Supported_Cameras
RawDigger also has a list of supported cameras and a similar disclaimer that it can open other files but they may not look "accurate". Again - I can open files in RawDigger that aren't on the supported list.
https://www.rawdigger.com/usermanual/cameralist
...so looks like a generic fallback sensor input profile than.
So - maybe "sensor aware" is an oversimplification - they will open files that aren't (yet) on the supported list.
But to say that the Sensor has no opinions about colors (how frequencies get measured, and then therefore turned into RGB values) is an oversimplification.
Hmmm technically you are correct but from the photographic point of view, everything starts from the moment when the image is created, so after demosaic is done. So to my opinion, this oversimplification yes, it do exist on my side, but it doesn't hurt given the discussed purpose
I've done some testing around this - you can read more here:
http://sodium.nyc/blog/2019/12/camera-color-science-does-it-exist-and-if-so-what-to-do-aboutit
http://sodium.nyc/blog/2019/12/camera-calibration-can-it-eliminate-differences-betweencameras
http://sodium.nyc/blog/2020/06/what-is-good-skin-tone-and-how-to-get-it
Thank you, it's very interesting, I'll read these later today.
Camera JPG Color Shootout - where you can look at images from a number of cameras and decide for yourself. Currently the X-Trans1 wins in both natural light & studio skin tone tests.
http://sodium.nyc/blog/2020/05/camera-jpg-portrait
Again I bet you may replicate this OOC JPEG look with any sensor given the proper sensor input profile, proper tone curve and proper LUT.
I typed up a large point by point response but decided summarize my points here to create a more coherent narrative.
I suspect that there may be information that's irretrievably "lost" at the CFA.* I haven't really put this to the test but people with the proper measuring tools may be able to prove/disprove this.
Phase One (for example) put a lot of effort into designing their Trichromatic sensor to have very good separation at the CFA. Other camera manufacturers purposefully design their sensors to have less color separation at the CFA. [edit: regardless of "manufacturer" - the camera brand designs the CFA]
Per one reviewer: the difference in practical terms is subtle. Importantly, however, it is there, and you can see it. Granted, it isn't like many of these differences in color rendition could be achieved via post-processing, but it would take time and some of the tonal transitions would be very difficult to replicate, particularly without sacrificing the integrity of the image file and quality.
While we can get cameras to produce color swatches after calibration, I question whether the subtle variations in between said swatches can be matched as well.
One frequency of light can be replicated by two or more different frequencies of light as far as the human eye is concerned - this is the basis of all imaging from paintings to CMYK print to RGB monitors and the basis of our whole discussion.
Given that - will two different sensors respond the same to a) the single frequency of light and b) two frequencies of light that the eye would interpret the same as a that single frequency? In way that the rest of the imaging pipeline would produce the same RGB values in both scenarios? Or at least produce consistency among brands?
My suspicion is that - maybe not. Maybe the fact that a Nikon camera's red sensels extends into the blue area means that it is in fact impossible for Nikon cameras to perfectly match Canon cameras, whose red sensels do not extend as far into the blue region and vice versa.
Even if you can get two cameras to reproduce a color chart perfectly, there is much more variation in the real world - two randomly selected pixels in the photo of an apple may have very different values on the sensor & whether or not you can get two cameras to produce the same RGB values for those pixels is IMO questionable. Close - sure, but identical? Maybe not. And when it comes to replicating skin tones, humans very attuned to those subtle variations.
Finally - while it's fun to debate this stuff online (I enjoy a good intellectual conversation), in the real world, I just don't have the time to bother with all this stuff. I much prefer to use the equipment I enjoy using and that gets me the results I like. I don't mind using the older, slower X-Trans1 cameras when I want the X-Trans1 colors.
Maybe the upgrades in speed, handling, megapixels, video capabilities etc. make the upgrade worthwhile (I do own 3 generations of X-Trans sensors and use them all), but editing can be time consuming enough without getting to a baseline that I like - RAW or JPG. I much prefer to be able to hand over JPGs to a client soon after the shoot than spend time in calibration. Even if it's just a one time cost per camera. If I shoot Nikon it's because I want Nikon colors or lenses or whatever. If I shoot Canon it's because I want Canon colors and so forth.
While I did enjoy calibrating all my equipment at one point in my life - for my purposes, I want to spend less time second guessing my editing choices (when it comes to skin tones it's very easy to go very wrong) and more time shooting or building my website or posting in internet forums.
If the camera produces JPGs I like - that's the dream. Zero work in post, just share the photos via dropbox.
Human vision behaves more like the "standard" sensor (with a red "hump" in the blue area and with less red sensitivity overall) which may be why Nikon chooses to have this "hump". Leica goes even further - though I like Nikon colors and dislike Leica colors.
Leica allegedly designed their M8 sensor to "look like Kodachrome" - and I suspect Nikon had a similar agenda.
* "lost" implies that there is an ideal, which there may be. But I do not mean it to imply that one camera captures more perfect data than another, rather that all design is compromise and each is tuned to the colors they want their cameras to produce.
Maybe a weaker red response means the red channel is less likely to blow out under certain circumstances and that's why a weaker red response is desired.
Maybe a bit of red response in the blue area mimics the eye better and produces more pleasing skin tones, but at the sacrifice of sky blues (Canon cameras are allegedly better at producing good sky blues).
IDK I'm not a camera engineer, just a photographer with too much time on a Saturday to post on internet forums....
