Maxpool layer

class layers.maxpool_layer.Maxpool_layer(size, stride=None, pad=False, input_shape=None, **kwargs)

Bases: NumPyNet.layers.base.BaseLayer

Maxpool layer

sizetuple or int

Size of the kernel to slide over the input image. If a tuple, it must contains two integers, (kx, ky). If a int, size = kx = ky.

stridetuple or int (default = None)

Represents the horizontal and vertical stride of the kernel (sx, sy). If None or 0, stride is assigned the same values as size.

input_shapetuple (default = None)

Input shape of the layer. The default value is used when the layer is part of a network.

padbool, (default = False)

If False the image is cut to fit the size and stride dimensions, if True the image is padded following keras SAME padding, see references for details.

>>> import os
>>>
>>> import pylab as plt
>>> from PIL import Image
>>>
>>> img_2_float = lambda im : ((im - im.min()) * (1. / (im.max() - im.min()) * 1.)).astype(float)
>>> float_2_img = lambda im : ((im - im.min()) * (1. / (im.max() - im.min()) * 255.)).astype(np.uint8)
>>>
>>> filename = os.path.join(os.path.dirname('__file__'), '..', '..', 'data', 'dog.jpg')
>>> inpt = np.asarray(Image.open(filename), dtype=float)
>>> inpt.setflags(write=1)
>>> inpt = img_2_float(inpt)
>>>
>>> inpt = np.expand_dims(inpt, axis=0)  # Add the batch shape.
>>> b, w, h, c = inpt.shape
>>>
>>> size = (3, 3)
>>> stride = (2, 2)
>>> pad = False
>>>
>>> layer = Maxpool_layer(input_shape=inpt.shape, size=size, stride=stride, padding=pad)
>>>
>>> # FORWARD
>>>
>>> layer.forward(inpt)
>>>
>>> forward_out = layer.output
>>>
>>> print(layer)
>>>
>>> # BACKWARD
>>>
>>> delta = np.zeros(inpt.shape, dtype=float)
>>> layer.delta = np.ones(layer.out_shape, dtype=float)
>>> layer.backward(delta)
>>>
>>> # Visualizations
>>>
>>> fig, (ax1, ax2, ax3) = plt.subplots(nrows=1, ncols=3, figsize=(10, 5))
>>> fig.subplots_adjust(left=0.1, right=0.95, top=0.95, bottom=0.15)
>>> fig.suptitle('MaxPool Layer

size{}, stride{}, padding{} ‘.format(size, stride, pad))

>>>
>>> ax1.imshow(float_2_img(inpt[0]))
>>> ax1.set_title('Original Image')
>>> ax1.axis('off')
>>>
>>> ax2.imshow(float_2_img(forward_out[0]))
>>> ax2.set_title('Forward')
>>> ax2.axis('off')
>>>
>>> ax3.imshow(float_2_img(delta[0]))
>>> ax3.set_title('Backward')
>>> ax3.axis('off')
>>>
>>> fig.tight_layout()
>>> plt.show()

• https://docs.scipy.org/doc/numpy/reference/generated/numpy.lib.stride_tricks.as_strided.html

• https://stackoverflow.com/questions/42463172/how-to-perform-max-mean-pooling-on-a-2d-array-using-numpy

• https://stackoverflow.com/questions/42463172/how-to-perform-max-mean-pooling-on-a-2d-array-using-numpys

backward(delta)

Backward function of maxpool layer: it access avery position where in the input image there’s a chosen maximum and add the correspondent self.delta value. Since we work with a ‘view’ of delta, the same pixel may appear more than one time, and an atomic acces to it’s value is needed to correctly modifiy it.

Parameters

delta (array-like) – Global delta to be backpropagated with shape (batch, out_w, out_h, out_c).

Return type

self

forward(inpt)

Forward function of the maxpool layer: It slides a kernel over every input image and return the maximum value of every sub-window. the function _asStride returns a view of the input arrary with shape (batch, out_w, out_h , c, kx, ky), where, for every image in the batch we have: out_w * out_h * c sub matrixes kx * ky, containing pixel values.

Parameters

inpt (array-like) – Input batch of images, with shape (batch, input_w, input_h, input_c).

Return type

self

property out_shape

Get the output shape

Returns

out_shape – Tuple as (batch, out_w, out_h, out_c)

Return type

tuple