Loading...

Depthwise-separable convolutions: how MobileNet gets ~8-9x fewer multiplies by splitting one conv into two | AIWedia