How to batch export image variations?

DT 3.8.1 on debian

After I have done my processing, I do the export with the: RL output sharpen.

The problem is it’s done after the resizing and it also depends on the image, I can’t preview, this means that I have to try one sigma at a time.

Can I do that at the command line, specifying the sigma for different raduises?

or

can I do it with gmic with different reduces (I have looked at the https://gmic.eu/reference/gmic_reference.pdf and can’t find rl sharpening)

Thanks

I’d look at the Diffuse or Sharpen preset for sharpening, either AA or non-AA. I’ve found this completely removes the need for post sharpening.

I did and I disagree, it’s very much a matter of image/subject. For photos of dog, it makes a big difference.

  1. resized to 1500px by dt
  2. view in geeqie
  3. left: sharpen/diffuse dehaze + sharpen demosaicing aa
  4. right: sharpen/diffuse dehaze + sharpen demosaicing aa + output sharpening.

When resized to 725px wide, the difference is even greater.

Thanks

Do you use high quality resampling in export? What are your interpolation settings? Did you enable any setting that may impair quality?

How many output sizes do you use?

Iirc, output sharpening is used to correct for the loss in perceived sharpness due to the down-sampling, and is not dependant on image content. So for a given camera and output size, the required sharpening should be constant.

That suggests you could define presets for your output sharpening and your export sizes.

Here are my settings:

  • high quality resampling: yes
  • interpolation: I don’t know, which one is it?

here’s my print screen

Screenshot_2022-02-20_11-18-37

Thanks

I usually use 1 down sampling size: 725x725. That’s 80% to 90% of the time. Then I occasionally use 1500x1500 and 2500x2500 or full size (0x0).

A preset would seem like a very good idea. Thanks

If you didn’t change it, it’s still probably the recommended default.
The interpolators can be adjusted in the preferences.
https://docs.darktable.org/usermanual/3.8/en/preferences-settings/processing/

It’s set to the default: lanczos3, but in the docs it says:

pixel interpolator (scaling)
lanczos3 can cause pixel overshoots leading to artefacts but sometimes gives a more crisp visual appearance. This option is therefore only provided for transforming (scaling) algorithms and is the default value.

It looks like the output sharpening richardson-lucie does give me the “crispiness”, short of crunching, then should I change the interpolator?

Thanks

If you don’t see artefacts, then there’s no reason to change it, I think.

Thank for taking the time to explain.