it's cool. but I really don't understand why RGB is the goto encoding, it's easy to get unwanted artifacts and longer training times. YCbCr is better imo, for

it's cool. but I really don't understand why RGB is the goto encoding, it's easy to get unwanted artifacts and longer training times. YCbCr is better imo, for some cases you can only use the Y channel for edge detection which is much faster and more compact. any ML engineers can shed some light on the topic? perhaps I'm missing something.