4 ms·
> Why have your first layers of CNNs relearn what's already smartly encoded in the JPEG bits in the first place before it's blown into a bloated height x width
by amelius 3y ago
> Why have your first layers of CNNs relearn what's already smartly encoded in the JPEG bits in the first place before it's blown into a bloated height x width x 3 float matrix?
Because data augmentation is much easier in the latter representation.
Also, if you rotate images as part of data augmentation, then that is already so expensive that any speedup from going directly to JPEG becomes negligible in comparison.