nn.functional.adaptive_max_pool3d

lucid.nn.functional.adaptive_max_pool3d(input_: Tensor, output_size: int | tuple[int, int, int]) → Tensor

Applies 3D adaptive max pooling over a volume input to reach a desired depth, height, and width.

Function Signature

def adaptive_max_pool3d(input_: Tensor, output_size: int | tuple[int, int, int]) -> Tensor

Parameters

• input_ (Tensor): The input tensor of shape \((N, C, D, H, W)\).

• output_size (int or tuple of int): Target output shape \((D_{out}, H_{out}, W_{out})\).

Returns

• Tensor: Tensor of shape \((N, C, D_{out}, H_{out}, W_{out})\).

Behavior

Kernel size and stride are computed in all three spatial dimensions to match the given output size, and max pooling is performed over each resulting subvolume.

Examples

from lucid.nn.functional import adaptive_max_pool3d

input_ = Tensor.ones((1, 1, 8, 8, 8))
output = adaptive_max_pool3d(input_, output_size=(2, 2, 2))
print(output.shape)  # (1, 1, 2, 2, 2)