Accelerated by another 5% using FP16/32 Batch-norm for Tensor Cores.

AlexeyAB

2018-04-16 9bae70b22549b68f5cdeece8b6c3b3de00c22714

 cfg/vgg-16.cfg

@@ -1,12 +1,11 @@
[net]
batch=128
subdivisions=8
subdivisions=4
height=256
width=256
channels=3
learning_rate=0.00001
momentum=0.9
seen=0
decay=0.0005

[crop]