dsplit

→list[Tensor]

dsplit(input: Tensor, indices_or_sections: _int | Sequence[_int])

Split a tensor along its third (depth) axis.

NumPy-style depth-wise split: cuts the input into pieces along axis 2. Mirrors vsplit (axis 0) and hsplit (axis 1).

inputTensor

Input tensor with at least 3 dimensions.

indices_or_sectionsint | Sequence[int]

int: number of equal-sized splits along axis 2; it must divide the axis length.
Sequence[int]: cut indices along axis 2.

list[Tensor]

Sub-tensors whose concatenation along axis 2 reproduces input.

lucid._C.engine.InvalidArgument

If input.ndim < 3, or a section count does not divide axis 2.

For a tensor of shape $(\dots, D, \dots)$ split into k pieces along axis 2, each piece has shape $(\dots, D/k, \dots)$ .

>>> import lucid
>>> x = lucid.arange(24).reshape(2, 3, 4)
>>> [s.shape for s in lucid.dsplit(x, 2)]
[(2, 3, 2), (2, 3, 2)]