[PlayRaw] Departing Storm


(Elle Stone) #1

This is my first “play raw”, and also possibly the first image I ever processed start-to-finish using high bit depth GIMP 2-9 built from git, back in 2013, using dcraw to output a scene-referred image file.

Back in 2013 I never managed to produce a final image that I actually liked. So all these years later I decided to try again using GIMP-2.10, this time using darktable to make a scene-referred rendition via the darktable GIMP plug-in:

080724-1822-100-1479.cr2 (9.0 MB)
Raw file licensed CC-NC-SA (https://creativecommons.org/licenses/by-nc-sa/3.0/).
080724-1822-100-1479.cr2.xmp (6.7 KB)
CameraRGB-elle-V4-g10.icc (892 Bytes)

The scene is the rapid departure of a violent rainstorm. Imho the image composition isn’t too awfully bad though I wish the capture had been shot with a slightly wider lens. But after interpolation, I found (and still find) it difficult to process the scene-referred image into a satisfying final image:

  • It’s a high dynamic range image, by which I mean that a reasonable placement of middle gray - for example put middle gray on the green grass in front of the white utility building - will push the RGB channel values of much of the cloud-covered portions of the sky well past 1.0f and in some places past 2.0f. There aren’t any blown pixels in the raw file because I exposed for the clouds. However, in darktable the white balance multipliers drove the red channel values of the darktable output above 1.0f in small portions of the brightest clouds.

  • I shot the scene through a green-tinted passenger-side front window of a car going maybe 45 mph down the road, just as the storm had come to an end. So portions of the lower right corner of the scene are distorted and obscured by rain drops and wind-blown debris that accumulated in the lower right corner of the window. Back in 2013 I made an effort to repair the damage, but this time I didn’t even try, wanting instead to concentrate on tone-mapping.

  • The camera white balance is “uniwb”. Normally for midday sunny-ish outdoor scenes in the raw processor I would just use Daylight white balance. But that wouldn’t remove the green tint from the car window. So I white-balanced the image on a white cloud.

After transferring the image from darktable to GIMP-2.10, I made several black and white renditions, but none of them sufficiently conveyed the impression that a storm is being pushed out of the scene. Here’s the one I liked the best (and I don’t like it very much):

To try to better capture the idea of a departing storm, I added the color information back in to the tone-mapped black and white rendering using GIMP’s “Luminance” blend mode over the base “from the raw processor” color layer. The resulting color image had a slight overall red color cast - the white portions of the clouds weren’t as neutral as they seemed when looking at the dull scene-referred image in the raw processor.

To remove the red color cast, I re-white-balanced the base color image on the roof of the white utility building (still using GIMP, but first converting the image back to the camera input profile). This made nicer sky and cloud colors, but the vegetation colors still weren’t right.

So I took the re-white-balanced color rendition into RawTherapee and used the CIECAM02 module to further modify the white balance using chromatic adaptation, and then continued working with the image to try to capture the sense that a storm had just ended. Here’s the result so far:

For both versions shown above, all “tone mapping” was done in GIMP-2.10, using masks and layers plus basic editing tools: Exposure compensation, Levels, Curves, and layer blend modes. I didn’t use any dedicated tone-mapping algorithms. This is the way I usually - well, always - process images, and usually I’m happy with the results.

But my two renderings of the departing storm clearly have a lot of room for improvement. So I’m hoping to see what other people might do perhaps using more sophisticated tone-mapping tools - there’s a lot of such tools out there - which to use? where to start?

Messed up colors by the number - darktable's new unbreak and filmic modules
(Thomas) #2

Thanks for sharing! darktable only.

080724-1822-100-1479.cr2.xmp (8.9 KB)


080724-1822-100-1479.cr2.pp3 (13.2 KB)
I’m not sure how to correctly use the .icc profile, so I’ve done the white balance with eye dropper and curves :neutral_face:

(dumb) #4

Now that there’s talk of storms, have an operation NOAA IR satellite image enhancement colour scheme:

…and of course, a lazy attempt at using G’MIC’s syntexturize among other things to whip up a very strong northern hemisphere tropical cyclone:

(Shreedhar Inamdar) #5

Very interesting image @Elle. Here is an attempt with RawTherapee dev 5.4-409-g99caa76f7. The bulk of the processing is via the Tone curve in the Exposure module and the LAB module.
080724-1822-100-147-1.jpg.out.pp3 (12.1 KB)

And an alternate take with bluish cast removed. (Don’t know which is more realistic)
080724-1822-100-147-2.jpg.out.pp3 (11.9 KB)


Thanks for sharing this nice photo. I don’t use a calibrated monitor but I mathematically measured different parts of the photo including sky, grass, leaves, clouds, the roof of the building, and roadway using darktable’s color picker module and Lab values for them seem fairly reasonable. The luminance value of the grass is ~44 which is in the right Zone (Zone IV) according the Zone system. Also I used the new “retouch” iop of darktable (latest git version) to remove the spots from the lower right part of the photo. All work was done in darktable.

darktable 2.5.0: 080724-1822-100-1479.cr2.xmp (8.7 KB)


080724-1822-100-1479.cr2.pp3 (12.8 KB)
Tried to make it angrier.

Using rawtherapee newlocallab with an oblong spot on the foreground area to brighten that which is darkened by lowering the brightness in general to reveal the undulating cloud base.

(Elle Stone) #8

Thank you everyone! I want to look at the various pp3/xmp files and see what algorithms different people used. It took me several days to arrive at something even half-way passable, and having a way to get to something nice a little quicker would be wonderful. Though perhaps the real problem is that I don’t really know what I’m looking for when dealing with this sort of image.

I think a common theme is that everyone wants the shadows to be deeper than they are in either of my renditions. @shreedhar’s two renditions are closest to what I did, but @shreedhar’s renditions look a lot nicer and I think the deeper shadows have a lot to do with that.

Because of all the issues with setting a white balance, what came from the camera is no guide for this particular image. Also, the default dcraw camera input profile produces fairly anemic blues with raw files from the Canon 400D, which is why I ended up making a custom camera input profile - the reds from that camera also are anemic using the default dcraw input profile, to the point where way back then a lot of people were making custom camera input profiles.

In the real world sky blue on a sunny day ranges considerably from green-blue to violet-blue. In my rendition the sky hue ranges from LCh hue=255 in the cloud shadows and the lower sky, to 265 in the darker upper portion of the sky. When dealing with this sort of “mangled color” raw file I think one’s aesthetic judgement coupled with actually looking at blue skies is about the only way to make color judgements.

This section of the handprint website has some nice comments on sky colors:


When interpolating the raw file, choose the custom camera input profile instead of the profile supplied by the raw processor, but only if you actually like the resulting colors better than the default-supplied camera input profile. The step that I mentioned about converting back to the camera input profile to white balance on the roof of the utility building was only necessary because I picked the wrong neutral object in the scene to use for white balancing, and I didn’t want to start over from scratch with a new rendering.

I think auto white balance also would work for this raw file - it’s the sort of scene that auto white balance algorithms work well with.

Ever since discovering that what I thought was a calibrated CRT monitor actually had a seriously too light black point (so every photo I edited had horribly, horribly crushed shadows that looked “just fine” on that particular CRT), I’ve viewed “what’s shown on my screen” with a grain of salt. So like yourself I measure LAB/LCh values and evaluate tonality based on “typical” Lightness values from various descriptions of the Zone system.

That doesn’t mean I slavishly place everything right where the Zone system indicates is “correct” - even the practitioners didn’t do that. But the Zone system is a really nice “by the numbers” reality check. Back when I used PhotoShop I had a set of layers set up that I could drag onto my layer stack to show what’s in each zone. I sort of miss that in GIMP. Do any of our free/libre raw processors have such functionality? I mean apart from LightZone (which uses Java? it uses something that I don’t like to have installed on my computer).

Thanks! for mentioning that iop - it looks like it did a really nice job on removing the water spots.


Totally agreed. I believe the Zone system (at least in digital photography) is meant to give a vision on the lightness (luminance values) of different things but one should not insist to put everything where the Zone system defines.

(Carmelo Dr Raw) #10

Hi @Elle, and thanks for this challenging image!

I have tried to achieve a dramatic, yet not fake look, using photoflow only. Here is my current result, obtained with a mixture of dynamic range compression, curves, tone mapping, exposure/gamma adjustments, and some slight vignetting:

080724-1822-100-1479.pfi (66.4 KB)


Darktable 080724-1822-100-1479.cr2.xmp (10.3 KB)


080724-1822-100-1479.cr2-b.pp3 (11.6 KB)
Second try with a little more brightness :slight_smile:

(Isaac Ullah) #13

To me, this type of shot calls for moody black and white. Here I just messed with the base curve (and original white balance), then used the monochrome tool to add a slight red filter for B&W conversion. Enhanced details in the clouds with the equalizer, added some dodging and burning on the road and trees with the levels tool and some drawn masks, and then added a touch of grain to add some life back to the sky (it was close to posterizing at this point). Here’s the jpg and xmp:

080724-1822-100-1479.cr2.xmp (9.8 KB)


thanks @Elle for after-strom, saluting fellow developeers

Photoflo. Took dumb’s spiral idea and HIRAM’s moodyness, trying to bring up the “vortex” created by lower clouds, lower left bush, road, middle-right trees.

080724-1822-100-1479_phf.tif.pfi.zip (4.3 KB)

Then gimp for gmic for a bit of 3dness BS; help from gradients, had to lower exposure down, was getting too bright

just to discover upon export some “hard leftover” also dots all over left corner, well … FUXKKKK!!! :see_no_evil: :sweat_drops: :beer: :zzz:

All is good; next time better, cheers

@msd I know what you did ,-)

(Glenn Butcher) #15

Your raw exposed a badness in my rawdata clipping, so I’ll be fixing that…

I did get it open raw, with your camera profile assigned. Here’s where I got it to:

Then, I got all artsy, exported a float tiff and opened it with the GIMP 2.10.7 AppImage:

Hmmm, was so intent on removing the power lines I forgot about the splotches in the blue sky…

(Shreedhar Inamdar) #16

Thank you @Elle for your comment and the reference. I am looking at it and trying to improve my understanding of colors.
Here is my thinking behind the submitted processing.

  1. It became clear to me that if the areas between the branches of the tree in the foreground becomes black, the balance of the image gets disturbed (it becomes bottom heavy). Therefore, all the processing is done so that those shadowy areas are never black.
  2. May be due to photo being taken through glass, the colors seem very muted and contrast was too low. Used the A and B channels of LAB to mildly improve the intensity of the colors.
  3. (As the tree in the foreground has to be kept with low contrast) the way to improve contrast is to switch on local contrast module and by locally adjust brightness difference. So used LH module of LAB Tab to make the green field in front of the white house dark and the brown field adjacent to it a bit light.
  4. From your description it was also clear that you are looking at the full DR in the photo. So used the Tone Curve and the A curve (in LAB) to make sure that the histogram is fully stretched from left to right.
  5. I used the sharpness and noise reduction (I saw that the blue expanse of the sky was speckled with noise so had to use NR) from the Wavelet module (just because I prefer them) to add further punch to the image.
    That was it. I found the image to be very delicately balanced. Both in terms of framing and color composition. So thought that doing anything more will ruin it. But of course, could not resist to add a very mild vignette to it.
    EDIT 6. I forgot to say that I used the Blue curve in the RGB module of the Color Tab to reduce the bluish tint for the second rendering.


RT version
The main work was done using RGB curves.
I didn’t do any noise reduction or sharpening.

My few thoughts:

  1. it shouldn’t be too bright or looses menace.
  2. it shouldn’t be too dark or too contrasty as I am going for a natural feel
  3. The colors of the vegetation shouldn’t be too juicy as there doesn’t seem to be direct sunlight anywhere and I wanted a washed look.
  4. Tried to pull the blue sky down as it just looked to happy…the aim was slightly gloomy.

080724-1822-100-1479_R.jpg.out.pp3 (11.3 KB)


darktable has a Zone system iop that highlights the corresponding areas when you hover the mouse pointer on a zone. Also you can use the parametric mask for L channel in combination with “display mask” functionality to highlight different zones. I use the latter in Tone curve iop. I have created 11 presets for this purpose.

(Thomas) #19

In contrast to my firdt try, a little bit more “rainy” and “washed out”.

080724-1822-100-1479.cr2.xmp (7.2 KB)

(Elle Stone) #20

@Thomas_Do - oh, I really like your rainy black and white rendition. Somehow you balanced the contrast in the sky so incredibly well with the contrast in the ground, and also kept the ground dark without it somehow looking out of balance with the sky.

Studying some of the other “darker ground” renderings such as @chroma_ghost’s along with @Isaac’s and @HIRAM’s very dark interpretations had already convinced me that a darker interpretation might be a nice way to push the image. But you did two things I never would have thought of doing: You pushed the bright areas of the sky quite a lot especially up at the top (I had worked hard to keep that area dark). And you managed to keep the ground areas somehow soft and rounded, maybe “organic” is the right word, without having it look mushy. It’s really easy with the ground area in this image to end with a rather crunchy look, I’m guessing from noise as that portion of the image is fairly underexposed.

The white frame you added I think is necessary to give context to the bright upper edge of the image. I’m looking at the post using pixls.us dark theme, so that might be why the frame looks so good, separates the dark image from the dark theme background and keeps the upper cloud along the edge from looking too bright.

I’ve started to download and explore the various xmp/pp3/pfi files to see what algorithms people are using. I’m really impressed with how easy it is to get a good rendition of the shadow tonalities from the various raw processors, and especially for the lighter, happier color versions of the image that include ground renditions that make the trees and field fairly light. If I had one complaint about using GIMP for this sort of image, it would be that modifying shadow tonality is just not very easy, and yet if the ground tonality isn’t modified commensurately with the sky, the whole image falls apart, as @shreedhar notes it’s a delicate balancing act .

Is there a way to rename a darktable xmp file such that darktable can open the raw file using "version A.xmp, “version B.xmp”, etc? Or a way to open the raw file and then load a different xmp file?