max_pool2d_op

→TensorImplPtr

int max_pool2d_op(const int & x, int KH, int KW, int stride_h, int stride_w, int pad_h, int pad_w)

source edit

Python wrappers

pylucid.nn.functional.max_pool2dfunction

Two-dimensional max pooling over a batch of feature maps.

Emits the maximum of each $KH \times KW$ sliding window. Provides local translation invariance and is the canonical downsampling primitive in convolutional classification networks.

Parameters

xTensorImplPtr

Input of shape (B, C, H, W).

KH, KWint

Per-axis window size.

stride_h, stride_wint= None

Per-axis stride. 0 is a sentinel meaning "use K". Default: 0.

pad_h, pad_wint= None

Per-axis zero-padding. Default: 0.

Returns

TensorImplPtr

Output of shape (B, C, H_out, W_out) with $H_\text{out} = \lfloor (H + 2 p_h - KH)/s_h + 1 \rfloor$ and analogously for $W_\text{out}$ .