Hello people,
I am trying to compute bit error rate as metric. Currently I am using numpy and more specific np.unpackbits(x)
I am wondering if there is a similar operation in pytorch to take advantage of gpu.
def __call__(self, output, target):
t_target = target.detach().cpu().numpy().astype(np.uint8)
t_output = output.detach().cpu().numpy().astype(np.uint8)
target_bits = np.unpackbits(t_target).reshape(
-1, 8)
output_bits = np.unpackbits(t_output).reshape(
-1, 8)
output_bits = output_bits.reshape(-1)
target_bits = target_bits.reshape(-1)
absolute = np.abs(
output_bits.astype(np.int) - target_bits.astype(np.int))
errors = np.sum(absolute)
return errors