nn.functional.adaptive_max_pool2d¶
Performs adaptive max pooling on a 2D input tensor to produce the specified output height and width.
Function Signature¶
def adaptive_max_pool2d(input_: Tensor, output_size: int | tuple[int, int]) -> Tensor
Parameters¶
input_ (Tensor): The input tensor of shape \((N, C, H, W)\).
output_size (int or tuple of int): Desired output size \((H_{out}, W_{out})\).
Returns¶
Tensor: Tensor of shape \((N, C, H_{out}, W_{out})\).
Behavior¶
Kernel size and stride are computed per spatial dimension to reduce the input to the given output size, with max pooling over each region.
Examples¶
from lucid.nn.functional import adaptive_max_pool2d
input_ = Tensor.ones((1, 1, 6, 6))
output = adaptive_max_pool2d(input_, output_size=(3, 3))
print(output.shape) # (1, 1, 3, 3)