Haven't had the chance to compare this method to converting it through python.

Do you think that doing the conversion on the cache side will be faster?

Thanks, increasing the buffer to 1000000 reduced the processing of a 18MB image from 20 seconds to 2-3 seconds.