Cutout#

class myoverse.transforms.Cutout(n_holes=1, length=50, p=0.5, dim='time', **kwargs)[source]#

Randomly zero out contiguous regions.

Parameters:

n_holes (int) – Number of regions to cut out.
length (int | float) – Length of each cutout. If float < 1, fraction of total length.
p (float) – Probability of applying cutout.
dim (str) – Dimension to cut along.

Examples

>>> x = torch.randn(64, 2048, device='cuda', names=('channel', 'time'))
>>> cutout = Cutout(n_holes=3, length=50)
>>> y = cutout(x)

Methods

`__init__`([n_holes, length, p, dim])
`_apply`(x)	Apply the transform.