Hey guys,
i am preparing a machine learning dataset with sony raw images.
I have several thousand images that are exactly the same lighting (flash) and background, only the subject (plants) change. I have not much experiance in processing raw images, i would like to have the noise, lense distorion, vingetting removed and good overall lighting. Also maybe its possible to do color correction with the color card in view?
This should be straight forward. Perform the adjustments on one image and apply the profiles to the whole batch and export. That may take a while with the number of images you have at hand…
It is great you have a color card. You may get all the neccessary scene parameters from there. Adjust the exposure and black point such, that the white patch is at 94% luminocity and the black one is at 4%. This is defined by the standard.
The lens corrections you just have to activate and the noise reduction should also work out-of-the box. If you want, you can actually use a custom color profile.
I personally use this software (link below), which takes the raw file and outputs a .dcp profile, which you can directly load into RawTherapee.
To keep things reproducible, you should also fix the gradation curve. If you are only doing computer vision, I reccomend a linear tone curve, which represents the true illumination. The only reason why we ususally use a non-linear mapping is because our eyes don’t percieve brightness linearliy.
I adjusted the exposure and black level, removed the luminance noise reduction. Any further post-processing step which requires low noise - e.g. edge detection - will do its own low-pass filtering.
I also deactivating the CA-correction. In your case, it is severely over-correcting and causing much larger fringes than your lens produces.