padding

mfai.pytorch.padding.pad_batch(batch, new_shape, mode='constant', pad_value=0)

Given batch of 2D or 3D data and a shape new_shape,

pads the tensor with the given pad_value.

Parameters:

• batch (Tensor) – the batch of values of shape (B, C, D, H, W) or (B, C, H, W).

• new_shape (Size) – Target shape (D, H, W) for 3D tensors or (H, W) for 2D tensors to be given.

• mode (Literal['constant', 'reflect', 'replicate', 'circular']) – the padding mode to be used. Defaults to “constant”. Same as torch.nn.functional.Pad

• pad_value (Optional[float]) – the padding value to be used. Defaults to 0.

Returns:

The padded tensor.

Return type:

Tensor

mfai.pytorch.padding.undo_padding(batch, old_shape, inplace=False)

Removes the padding added by pad_batch.

Parameters:

• batch (Tensor) – The padded batch of data

• old_shape (torch.Size) – The original shape of the data

• inplace (bool) – Whether the returned tensor is just a sliced view of the given tensor or a new copy.

Return type:

Tensor