def max_pool(value, ksize, strides, pads=(0, 0, 0, 0),
padding=None, data_format="NCHW", name=None):
"""
Performs the max pooling on the input.
Args:
value: A 4-D `Tensor` with type `tf.float32`.
ksize: A list of ints that has length 4.
strides: A list of ints that has length 4.
pads: A list of ints or a int.
padding: A string, either `'VALID'` or `'SAME'`. (deprecated)
data_format: A string. 'NHWC' and 'NCHW' are supported.
name: Optional name for the operation.
Returns:
A `Tensor` with type `tf.float32`. The max pooled output tensor.
"""
if len(strides) != 4:
raise ValueError('strides must be a list of length 4.')
if len(ksize) != 4:
raise ValueError('strides must be a list of length 4.')
if data_format == 'NCHW':
if pads is None: pads = 0
return ops.Pool2D(value,
kernel_size=ksize[2:],
stride=strides[2:],
pad=pads,
mode='MAX_POOLING')
else: raise NotImplementedError()
def max_pool(value,
ksize,
strides,
pads=(0, 0, 0, 0),
padding=None,
data_format="NCHW",
name=None):
if len(strides) != 4:
raise ValueError('strides must be a list of length 4.')
if len(ksize) != 4:
raise ValueError('strides must be a list of length 4.')
if data_format == 'NCHW':
if pads is None: pads = 0
return ops.Pool2D(value,
kernel_size=ksize[2:],
stride=strides[2:],
pad=pads,
mode='MAX_POOLING')
else:
raise NotImplementedError()
def Setup(self, bottom):
input = bottom[0] if isinstance(bottom, list) else bottom
super(PoolingLayer, self).Setup(bottom)
return ops.Pool2D(input, **self._param)