Advice on processing steps with rawpy

[mbrown-gh]

Hello,

As mentioned here, I am interested in processing DNGs from RPi cameras (either V2 or HQ), with the ultimate goal of determining the average color of the image. I'm hoping to perform some of the steps listed here, and after completing this tutorial think I know how to use rawpy to do so. A few initial questions below:

• Does rawpy.imread modify the underlying data at all, or does it just load the image?

• Does rawpy subtract black levels (and if so, where does this occur)? Are the black levels automatically determined (e.g. from the DNG metadata), or do they need to be manually supplied?

• Especially for the RPi V2 camera, it seems likely I will need to perform a lens shading correction. Do you have any advice on how that step could be incorporated (e.g. with lensfunpy - I'm uncertain if RPi cameras are supported)?

Thanks!
Mike

[letmaik]

Calling rawpy.imread() and accessing .raw_image leads only to calling libraw's open_file() and unpack(), see also https://www.libraw.org/docs/API-CXX.html#unpack.
Note that the docs mention "Data reading is sometimes (not frequently) affected by settings made in imgdata.params". See https://www.libraw.org/docs/API-notes.html#imgdata_params for details. rawpy does not allow to modify those settings prior to loading, which is an intentional restriction to keep the Python API simple since this is very rarely of use. I think in your case this shouldn't make a difference.
For further questions on how the raw file data relates to the data exposed by open_file()/unpack() you'd have to ask in the libraw forums. In general, the idea is that this is as close to the RAW data as possible, but I'm not too familiar with all the details of the libraw codebase.

Regarding metadata read from the file (or computed) by libraw, see https://letmaik.github.io/rawpy/api/rawpy.RawPy.html, e.g. black_level_per_channel. You can also specifiy a custom black level when using postprocess(), see https://letmaik.github.io/rawpy/api/rawpy.Params.html, but I think you don't want to go there and probably operate only on the raw data.

Vignetting correction (lens shading) is indeed supported by lensfunpy (or rather the underlying lensfun library), but it looks like there are no profiles available for RPi: https://wilson.bronger.org/lensfun_coverage.html. You can create your own profiles with manual calibration (https://wilson.bronger.org/lens_calibration_tutorial/) but this may possibly be too much work.