D700_20090529_9255.jpg.out.pp3 (12.0 KB)
D700_20090529_9255.NEF (12.4 MB)
This file is licensed Creative Commons, By-Attribution, Share-Alike
Here is my take, using Darktable 3.0rc. To tell you the truth, I rather like the original, soft feeling one more.D700_20090529_9255.NEF.xmp (12.2 KB)
Nice exposure, no clipping…
I did this on my 2core 2Gb backup machine in the basement; need to upgrade that thing…
group: (standard processing, 'cept I removed the subtract as the D700 doesn’t have a value)
- demosaic:amaze; (@heckflosse, I don’t see the artifacts we’re chasing in this image)
- tone:filmic,6.20,0.50,1.70,0.06,1.00 (This is the standard filmic curve, which yields a flat image. @age, I’m working on your premise here…)
- curve:rgb,0.0,0.0,54.0,9.0,122.0,63.4,255.0,255.0 (…with a S-shaped tone curve. Well, not too S-shaped; I just pulled down the lower part, let the upper part climb gradually to the top to keep definition in the highlights. Oh, to that end in the filmic operator, I unchecked ‘norm’, which scales the filmic function up to 1.0, so out of filmic the data only goes to about 0.84…)
- sharpen:0.5; (I’m improving USM sharpen; it now takes fractional values to build the kernel, and the UI actually shows you the kernel)
For sharpen = 0.5, the kernel is:
0.00 -0.12 0.00 -0.12 1.50 -0.12 0.00 -0.12 0.00
What lens is this?
Nice Seagull! I like the square crop …
Hmm, RT uses this kernel for sigma 0.5
0.0113437 0.0838195 0.0113437 0.0838195 0.6193470 0.0838195 0.0113437 0.0838195 0.0113437
D700_20090529_9255.NEF.xmp (10.5 KB)
I’ve used the base curve module in a way that it mimics exposure compensation with roll-off in the highlights before the real camera base curve.
I’ve then used the mystic golden spirales rules for the crop
Sorry, this is my version (darktable 3.0):
D700_20090529_9255.NEF.xmp (7.1 KB)
In some edits, the gull has a dark or light halo. I haven’t looked at the raw file yet; maybe the light soft halo was there to begin with.
Look at the raw file. If there is some halo, it maybe is caused by shooting at F1.8
Way back when I implemented this, I took the example kernel from here:
and worked up the equation that yielded the ratio between the center nymber and the outliers. The code in rawproc:
Lord knows it’s probably wrong, fruits of bear-of-little-brain…
Yes, I saw it in the background around the seagull’s head. I used a parametric mask on the background and reduced exposure there and it was gone
I just hard-coded this matrix in rawproc, and short-circuited the kernel constructor in favor of it. It blurs the image…
Shouldn’t it? It’s a gaussian blur kernel.
Y’know, now that I recall, mine is not USM, which requires a blur. Mine is straight convolution sharpening…