AdaptiveMaxPool2d

class torch.nn.AdaptiveMaxPool2d(output_size, return_indices=False)

Applies a 2D adaptive max pooling over an input signal composed of several input planes.

The output is of size $H_{out} \times W_{out}$, for any input size. The number of output features is equal to the number of input planes.

Each output element is the maximum over a rectangular region of the input. The bounds are computed independently for each spatial dimension, from an output index $k$, an input size $I$ and an output size $O$:

$$\text{start}(k, I, O) = \left\lfloor \frac{k \times I}{O} \right\rfloor \qquad \text{end}(k, I, O) = \left\lceil \frac{(k + 1) \times I}{O} \right\rceil$$

Writing $h_0 = \text{start}(h, H_{in}, H_{out})$ and $h_1 = \text{end}(h, H_{in}, H_{out})$, and likewise $w_0$ and $w_1$ from $(W_{in}, W_{out})$, the output is

$$out(N_i, C_j, h, w) = \max_{h_0 \le m < h_1} \max_{w_0 \le n < w_1} input(N_i, C_j, m, n)$$

When an input size is not divisible by the corresponding output size, these regions may have different sizes and may overlap. For example, a $3 \times 3$ input pooled to $2 \times 2$ takes the maximum over four overlapping $2 \times 2$ regions.

Parameters:

• output_size (int | None | tuple[int | None, int | None]) – the target output size of the image of the form $H_{out} \times W_{out}$. Can be a tuple $(H_{out}, W_{out})$ or a single $H_{out}$ for a square image $H_{out} \times H_{out}$. $H_{out}$ and $W_{out}$ can be either an int, or None which means the size will be the same as that of the input.

• return_indices (bool) – if True, will return the indices along with the outputs. Useful to pass to nn.MaxUnpool2d. Default: False

Shape:

• Input: $(N, C, H_{in}, W_{in})$ or $(C, H_{in}, W_{in})$.

• Output: $(N, C, H_{out}, W_{out})$ or $(C, H_{out}, W_{out})$, where $(H_{out}, W_{out})=\text{output\_size}$.

Examples

>>> # target output size of 5x7
>>> m = nn.AdaptiveMaxPool2d((5, 7))
>>> input = torch.randn(1, 64, 8, 9)
>>> output = m(input)
>>> # target output size of 7x7 (square)
>>> m = nn.AdaptiveMaxPool2d(7)
>>> input = torch.randn(1, 64, 10, 9)
>>> output = m(input)
>>> # target output size of 10x7
>>> m = nn.AdaptiveMaxPool2d((None, 7))
>>> input = torch.randn(1, 64, 10, 9)
>>> output = m(input)

forward(input)

Runs the forward pass.