No answer.
I ended up using a conversion to numpy array and back:
def augment_monochrome(x,
joint_transform,
noise_ampl=.2,
opencv_gpu_fix=True):
print('augment_monochrome call', end='...')
if opencv_gpu_fix:
x_aug0 = x.as_in_context(mx.cpu())
else:
x_aug0 = x
x_aug = F.repeat(x_aug0, repeats=3, axis=1)
x_aug_ = F.stack(*[F.swapaxes(
(1 + noise_ampl * np.random.normal()) * joint_transform(
F.swapaxes(x_aug[i, ...], 0, 2)
) + noise_ampl * np.random.normal(), 0, 2
) for i in range(batch_size)])
if opencv_gpu_fix:
x_aug = x_aug_.as_in_context(ctx)
else:
x_aug = x_aug_
x_aug = x_aug[:, :1, :, :]
x_aug = F.clip(x_aug, 0., 1.)
print('augment_monochrome finish', end='...')
return x_aug
slow, but it's honest work...
---
[Visit
Topic](https://discuss.mxnet.apache.org/t/attempting-to-use-augmentation-during-training/7044/2)
or reply to this email to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.mxnet.apache.org/email/unsubscribe/f93425b502c533820be94627ed77645f27b89875ad94fa8d6d14e96d7a9fd70a).
