MixUp#

class torchaug.transforms.MixUp(alpha, inplace=False)[source]#

Mix input tensor with linear interpolation drawn according a Beta law.

The shape of the tensors is expected to be [B, …] with … any number of dimensions. The tensor should be float.

Note

The tensor is rolled according its first dimension and mixed with one drawn interpolation parameter for the whole tensor.

Parameters:

  • alpha (float) – Parameter for the Beta law.

  • inplace (bool, optional) – Whether to perform the operation inplace.

    Default: False

forward(tensor, labels=None)[source]#

Mix the input tensor and labels.

Parameters:

  • tensor (Tensor) – The tensor to mix.

  • labels (Optional[Tensor], optional) – If not None, the labels to mix.

    Default: None

Returns:

  • mixed tensor.

  • mixed labels or None.

  • mixing coefficient.

Return type:

Tuple