we would like to record camera frames in YUV420 format and later turn them into RGB(A) to train networks.
When the networks run on the Drive PX we would use the dwImageFormatConverter to turn them into RGBA.
In order to minimize differences between training and evaluation, I’d like to exactly reproduce what the dwImageFormatConverter does when converting from YUV420 to RGBA.
I’ve used opencv cv2.cvtColor(img, cv2.COLOR_YUV2BGR_YV12), but the results are slightly different.
Obviously there are many different ways how this conversion could be done.
Is the algorithm used by dwImageFormatConverter documented?
Or is there a better way to do this?