recover ooc jpeg detail/sharpness from raw with darktable

I am trying to recover the same level of detail from RAW files using Darktable 3.8 that my Canon G7X mark II provides with JPEGs.

(I am aware that cameras occasionally overcook images, I just want to learn how to do it so that I can use it selectively.)

I attach a typical example image, JPEG has a lot more detail for the cat’s hair.


IMG_1946.CR2 (26.6 MB)

(EDIT: images licensed CC BY SA)

Also, generally, should I be using

  1. contrast equalizer,
  2. local contrast, or
  3. diffuse or sharpen (add local contrast preset),

or a combination of these? I am a bit confused about when to prefer one or the other.

1 Like

Hi there, @Tamas_Papp,

As per your wish: a swift, overcooked cat…


IMG_1946.CR2.xmp (10.1 KB)

Have fun!
Claes in Lund, Sweden

PS: There still is room for a few additional oompfs™:


IMG_1946.CR2.xmp (10.8 KB)

Thanks, but in case it is not clear: I want to reproduce the JPEG’s detail, more or less, not generate uncanny images with artifacts.

If you license the image properly and tag it as a play raw I’m sure you’ll get lots of alternatives!

IMG_1946.CR2.xmp (13.3 KB)

IMG_1946.CR2.pp3 (14.3 KB)

1 Like

one problem is that the focus seems to be on the ground at the cat’s feet and in front of it :wink:

1 Like

I used rawproc to process your raw to the linear RGB, to see what detail was available in the raw:

I don’t know if the screenshot will adequately show it, but the image’s focus plane is clearly not on the cat. Even if it was, f3.2 won’t help much with the whole cat.

So, detail in an image is about contrast and edge definition, and there are a number of ways to enhance both. The two obvious tool types are, aptly named, contrast and sharpen, but color saturation can also play a part. The cat has some nice gradation in its reddish-brown fur, so dialing up color saturation a bit can also help make things stand out.

So, right after the blackwhitepoint tool, I put in just a scooch of HSL saturation and convolution sharpening, and then a light filmic curve, in the order given here. Here’s the web-sized rendition:

1 Like


cat.sharpen.contrast.xmp (11.7 KB)

The cat not being in focus makes this harder than it needs to be :sweat_smile:

2 Likes

The look you see in your ooc JPG that you perceive as sharpness is really a compilation of tone mapping local contrast and sharpening. It’s the manufacturers secret sauce and why your camera costs so much.

You can try the details threshold slider with masking to try to focus and enhance your sharpening. Basically you will simply have to experiment to find a combination of tools that provides you with something close to what you are looking for. Sharpness is very very subjective…some people need it to hit them in the face and others see a light touch as the right amount.

Also you for sure need to assess it at 100 percent…Sometimes JPG images look better on first glance at a preview but when you zoom to 100 percent and compare to the raw edit…the JPG falls apart…

1 Like

For sharpening always use diffuse or sharpen module. It has been specially developed for this purpose. And it works in scene reffered workspace.

For the example above with the cat, just use two instances one after the other: first the preset lens deblur hard and then sharpen demosaicing (AAfilter).

Don’t forget to turn on the denoise (profiled) module first, as sharpening amplifies noise.

6 Likes

I have never used sharping yet, but came to the same conclusion as s7habo. Used medium instead of hard lens deblur preset.

DT 3.9 IMG_1946.CR2.xmp (10.4 KB)

1 Like

I modified the image license. Apologies about the focus (cats move fast!).

Thanks for all the edits and suggestions, I learned a lot.

@s7habo: Indeed Diffuse & sharpen sharpen + deblur fast works neat. So in Darktable 3.8 Local contrast and Contrast equalizer are kind of redundant and should not be used?

Sorry if this is a dumb question, but how do I “add” the xmp files people post here in lighttable so that they appear as edits for a duplicate?

In lighttable view select the image and press ctrl-d. You now have a duplicate. Apply the downloaded xmp (history stack->load sidecar file).

2 Likes

grafik

2 Likes

Exactly. These can still be used but you can expect artifacts when working in (the recommended) scene-reffered workflow.

2 Likes

Y’know, that’s the only challenge I see to an otherwise fine image - composition and exposure are quite nice.

To that end, it amazes me how recent cameras have made such a tizzy about finding eyes in autofocus. Doesn’t much help us poor landscape photographers… :crazy_face:

1 Like

Local contrast and unsharp mask could be used in the scene referred workflow

Should I append or overwrite?

Overwrite it if you have a fresh duplicate. That’ll apply just what’s in the xmp.

Appdend will add what’s in the xmp to whatever is already there/your default settings.

1 Like

In general you probably want to use overwrite, this makes sure that nothing stays behind from previous edits (a copy would include this).

Append would put all the new stuff on top of what’s already there. Not what you want in this case.

Too late… :slight_smile:

1 Like