As usual, there is a wide variety of interpretations of a PlayRaw image. Not counting the collage, there are almost 30 different edits here already, ranging from pretty realistic looking to aggressively enhanced colors to lightened pastels. @arctic, I’m curious, which one most closely resembles what you saw without a camera in front of your face? Not that this should be the yardstick for judging any particular edit…strictly curiosity.
I remember that I starred for a long time at that sunset. It was very orangy, and there wasn’t much purple. It felt also very bright, intense, and saturated.
Before taking the picture I was thinking “How will I able to make it feel like this in post?”.
Looking at all your amazing entries, I think the best matches with my perception are @hanatos (vkdt), @aurelienpierre (filmic+tone eq), @pass712 (sigmoid). They all sacrifice some saturation and contrast compared to the other entries but they are much more realistic. I guess that to fit such a high dynamic range in luminosity and saturation we need compromises, otherwise unrealistic results and limitations of the viewing device will damage the picture.
In my edits, I think I was biased from the Fuji SOOC JPG, which is contrasty and with a strange WB. Also, I overdid in saturation because I remembered the scenery so bright and colorful. I guess that the limitation of our screens should also constraints the choices during the raw postprocessing.
I wonder If that would still be the case with a 2000 nits OLED HDR screen.
Thanks for claryfying. Funny thing: I showed both of my edits to my wife and guess what she said…
Nice to mention my edit in such an illustrious circle even if the merits should go to @aurelienpierre who set the direction.
As usually so many interpretations of 1 source image. It was great to see all options!
It’s not just a matter of dynamic range of the destination space, it’s rather that any RGB gamut is pointy at the both ends, so as you reach 100% lightness, you necessarily converge to 0% chroma:
(from The sRGB book of color - Aurélien PIERRE Engineering)
Here, you see there is no really bright and really colorful orangy-red available in gamut. Actually, if we redraw that on a (hue, chroma) plane at 90% lightness, we get:
So, what usually happens without chroma mapping is the orangy-red you try to represent in the destination space at high chroma and high lightness gets clipped (at non-constant hue) to the closest color at similar chroma… which is yellow. That’s why most sunsets have that ugly hue break with rat-piss yellow in the middle of the red.
So if you want to preserve hue and lightness for bright areas, you need to reduce the chroma (what’s improperly called “desaturating”). Otherwise, you really need to darken the areas (with burning) that you want to keep colorful.
Notice this is shown here in sRGB, but larger spaces will exhibit the same behaviour and the same gamut shape, the difference being essentially the scaling (so chroma might spread wider in large gamut, and lightness might spread higher in HDR spaces, but the overall concepts remain).
DSCF9668.pfi (36.8 KB)
So I guess that the “issue” is RGB colorspaces/modeling of color. If we had a device able to capture and then send again the exact photons that we received in the retina while we were seeing the scenery, then we would have the same visual experience.
As far as I understood this cannot happen because we are wasting information in-camera, and our screens cannot display all the colors (they cannot display arbitrary spectra, but just a linear combination of RGB primaries spectra).
Is this shape caused by the limitations of the RGB modeling? Do you have any good references (book/papers) where to read about it? Thanks!
The reference is me… I simply plotted a JzAzBz gradient, discretized it every x %, converted it to sRGB and cut all patches that didn’t fit inside sRGB (aka one of the RGB coordinate was outside [0;1] once projected to display-referred). The shape is what it is when projected into perceptual, there is no design here, it’s pure data.
It’s not just what cameras capture, it’s also what LED screens can emit and paper can print, convolved with good old metameric error of tristimulus-based representation of light (aka different spectra can get encoded with the same tristimulus).
Take paper… the max luminance it can display is when it’s naked. Adding pigments add colors but also removes light, so more saturated (in pigments) induces darker. Aside from RGB encoding, there is no way to get colorful stuff at on reflective media at high luminance, because adding color = adding pigments = removing light.
Colorful at high luminance is laser light (monochromatic). Unfortunately, your good old LED displays are nothing like monochromators, their primaries have a rather large spectrum. But then, you make white (and luminance) as a by-product of adding R, G, and B. So, high luminance means high R and G and B, but then high chroma (colorfulness) implies high R or G or B (or ideally, laser… in any case : monochromatic-ish light). That’s an impossible situation.
It’s not so much a problem of data representation (we have ways of infering spectrum from RGB using LUTs and all), rather than a stupid problem of displaying technologies.
I think this scene offers a lot and if you are going for “realism” the subdued edits are likely best. With so much going on you may wish to focus say on the reflection in the water and the birds or the boat or whatever. That might change things on how you shape the light… but also when you have all that going on in the sky why not go a bit crazy just to see what shakes out… intent is really the decision maker. I like the more muted edits for their sense or realism but I also like the heavy contrast and crazy colors in some of them just for the pure visual entertainment…
With images like this, I’ve been messing with the two-curve approach: 1) loggamma to lift the shadows from the basement, and 2) a control point curve to shape the contrast and “toe-out” the bottom. Here’s what I got, in rawproc:
I don’t usually post twice to PlayRaw, but I think this is instructive…
I still wasn’t satisfied with the lift of the deepest shadows in my first attempt, so I played with it some more. Particularly, what I decided was that I could live with the lightness in the foliage if I just crushed the lowest parts to make some contrast. So, I just dragged the low-left point of the curve to 8,0 and did a little scooching around with the lowest part. That, and a touch of HSL saturation, and I get this, rawproc screenshot so you can see the curve:
I think your first one is much better …again my opinion. This has a nice pastel look when zoomed out, however again IMO the light on the boat looks way off and really washed out to me and the shadows now are laced with noise and what appears to be CA as well…I vote for your first one…
DSCF9668.RAF.xmp (7.9 KB)
I liked yteaot’s version as a BW. (Edit: liked it in color too.)
Added lens correction and a retouch, and then a luminance BW conversion.
DSCF9668_17.RAF.xmp (11.6 KB)
Welome…enjoy the discussions…
I find it interesting that the DR400 underexposed the image by 1 stop in terms of raw clipping… Not that it really affects anything, since it’s a pretty clean sensor to begin with.
- Profiled CA on
- Exposure Comp -1/3
- Highlight Rolloff Point 1.0
- Film Area 2232.4 to enhance finer details
- Drama 100
- Overdrive On for maximum punch
- White Clipping Point 0.466 to overall brighten the image
- Shadow Brightness 479.8 to brighten the shadows more
- Highlight Brightness 1000 to lift things even further
And a light application of experimental nonlocal means noise reduction.
I don’t know if it is the noise reduction but it has a oily quality to it (in particular, the water).
Actually, I think that’s JPEG compression. Let me try to downsize it less.