When using a 16 bit pixel format and supplying getNextNumPyFrame a np.zeros((2048, 2048), dtype=np.uint16) frame buffer, pixelink thinks that the array is too small. Supplying np.zeros((2048,2048*2), dtype=np.uint8) like used by the examples works, but makes less sense to me.
I thus added the array element bytesize to the buffer size calculation.

I'm aware that afterwards I've to call numpy's byteswap function for big/little endian conversion.