Conversation

Sebastian Krzyszkowiak

No manual editing this time, but an automatic postprocessing script using darktable-cli and rudimentary lens corrections. Takes ~30 sec to develop on the phone - about ten times longer than the default script used by Millipixels, but with much better results.

4
2
1

@dos link to script please :)

1
0
0

@dos /me stets an alarm for some time in the future :)

1
0
0

@joao There's still one "auto" button that I can't figure out how to press from the CLI interface 😭

1
0
0

@dos I was kidding about my alarm clock :)

0
0
0

@dos that's amazing! How did you achieve this?

1
0
0

@badumtss There wasn't much to "achieve" there, just a matter of sitting down and spending some time in front of darktable to come up with a useful preset. The real achievement yet to be made is to make it look close enough with at most 2-3 sec processing time on the phone.

1
0
0

@dos @badumtss is all the processing done on the CPU? Are there reasonable ways to offload to the GPU instead?

1
0
0

@pak0st @badumtss CPU. Darktable can use OpenCL, but I can't really tell how useful would that be on GC7000L even if it was actually implemented by etnaviv.

1
0
0
@dos @pak0st @badumtss Yes, you can do debayer (etc) on gpu. git@gitlab.com:tui/debayer-gpu.git But nicely integrating that is not easy; it should probably go into libcamera.
0
0
0

before / after (both fully automatic, just taking more time to process)

1
0
0

something dark...

1
0
0

and another cat, there's never too many cats

1
0
0

ok, that's enough varying lightning conditions (for now 😜)

0
0
0

@dos did you add the lens correction metadata to the dng or is it hardcoded in the postprocessing script?

1
0
0

@martijnbraam None of the above; I'm using Lensfun. I'll add it to the database once I'm reasonably convinced that the data is right.

1
0
0

@dos oh interesting option. I think it would be great to have some of the lens calibration in the DNG files so it works on all standards compliant parsers, not just the ones that have lensfun support.

0
0
1