nn.AdaptiveMaxPool2d¶

class lucid.nn.AdaptiveMaxPool2d(output_size: int | tuple[int, int])¶

The AdaptiveMaxPool2d applies adaptive max pooling on a 2D input, computing appropriate pooling parameters to reach the target output shape.

Class Signature¶

class AdaptiveMaxPool2d(nn.Module):
    def __init__(self, output_size: int | tuple[int, int])

Parameters¶

output_size (int or tuple of int): The target output spatial size \((H_{out}, W_{out})\).

Attributes¶

output_size (tuple of int): Stored target output size after adaptation.

Forward Calculation¶

Takes input of shape \((N, C, H, W)\) and produces output of shape \((N, C, H_{out}, W_{out})\). Kernel size and stride are computed automatically.

Behavior¶

Wraps adaptive_max_pool2d, computing parameters that evenly partition the input height and width into the target output dimensions.

Examples¶

import lucid.nn as nn

input_ = Tensor.ones((1, 1, 6, 6))
pool = nn.AdaptiveMaxPool2d(output_size=(3, 3))
output = pool(input_)
print(output.shape)  # (1, 1, 3, 3)