nn.AdaptiveMaxPool3d¶

class lucid.nn.AdaptiveMaxPool3d(output_size: int | tuple[int, int, int])¶

The AdaptiveMaxPool3d performs adaptive max pooling over 3D volumetric input, automatically adjusting kernel size and stride to achieve the specified output dimensions.

Class Signature¶

class AdaptiveMaxPool3d(nn.Module):
    def __init__(self, output_size: int | tuple[int, int, int])

Parameters¶

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

Attributes¶

output_size (tuple of int): Stores the output spatial size as a 3-element tuple.

Forward Calculation¶

Transforms input of shape \((N, C, D, H, W)\) into output of shape \((N, C, D_{out}, H_{out}, W_{out})\), adapting pooling parameters per dimension.

Behavior¶

Encapsulates adaptive_max_pool3d and dynamically computes pooling regions to match the given output volume dimensions.

Examples¶

import lucid.nn as nn

input_ = Tensor.ones((1, 1, 8, 8, 8))
pool = nn.AdaptiveMaxPool3d(output_size=(2, 2, 2))
output = pool(input_)
print(output.shape)  # (1, 1, 2, 2, 2)