Hello everyone,
new member here. I am using Libraw (more specifically Rawpy) for an application I am prototyping and I would like to speed up the raw processing as my application processes frames in two stages, first opening images with very conservative quality settings, performing all sorts of analysis and then reopening the same images in the background with high quality demosaicking.
For the initial rendition of the image I really don't need more than a "big thumbnail" (~1200x600px, i.e. ~1mp) and I was wondering how feasible it would be to modify the "half_size" parameter for Bayer-pattern files to be numeric and support powers of two greater than 1, i.e. discarding/averaging more RAW readings early in the processing stage.
I would imagine I can't use the thumbnails to this end, as my understanding is that these have migh have in-camera processing applied and I am working off as-raw-as-possible sensor ADU.
I am no stranger to C/C++ and had a quick look at the code but I was wondering whether something like this has been attempted before, as in this day and age of ever denser pixel-pitches it seems reasonable that others could benefit from such a feature.
I would further like to ask whether more experienced users/developers see any obvious roadblocks to implementing such a feature as my naive intuition would be that this only requires the sampling stride to be modified at the raw unpacking stage.
Any feedback is greatly appreciated,
m.