def avg_pooling(x, pool_h, pool_w, stride):
"""
Applies average pooling over an input array.
Args:
x (numpy.ndarray): The input array to be average pooled.
pool_h (int): Height of the pooling window.
pool_w (int): Width of the pooling window.
stride (int): The stride of the sliding window.
Returns:
numpy.ndarray, an output array after applying average pooling on input array.
"""
validator.check_integer("stride", stride, 0, Rel.GT)
num, channel, height, width = x.shape
out_h = (height - pool_h) // stride + 1
out_w = (width - pool_w) // stride + 1
col = im2col(x, pool_h, pool_w, stride)
col = col.reshape(-1, pool_h * pool_w)
out = np.mean(col, axis=1)
out = out.reshape((num, out_h, out_w, channel)).transpose(0, 3, 1, 2)
return out
def __init__(self, kernel_size, stride, pad_mode):
name = self.__class__.__name__
super(_PoolNd, self).__init__()
validator.check_type('kernel_size', kernel_size, [int, tuple])
validator.check_type('stride', stride, [int, tuple])
self.pad_mode = validator.check_string('pad_mode', pad_mode.upper(),
['VALID', 'SAME'])
if isinstance(kernel_size, int):
validator.check_integer("kernel_size", kernel_size, 1, Rel.GE)
else:
if (len(kernel_size) != 2 or (not isinstance(kernel_size[0], int))
or (not isinstance(kernel_size[1], int))
or kernel_size[0] <= 0 or kernel_size[1] <= 0):
raise ValueError(
f'The kernel_size passed to cell {name} should be an positive int number or'
f'a tuple of two positive int numbers, but got {kernel_size}'
)
self.kernel_size = kernel_size
if isinstance(stride, int):
validator.check_integer("stride", stride, 1, Rel.GE)
else:
if (len(stride) != 2 or (not isinstance(stride[0], int))
or (not isinstance(stride[1], int)) or stride[0] <= 0
or stride[1] <= 0):
raise ValueError(
f'The stride passed to cell {name} should be an positive int number or'
f'a tuple of two positive int numbers, but got {stride}')
self.stride = stride
def max_pool_with_argmax(x, pool_h, pool_w, stride, pad):
"""Max pooling with argmax."""
validator.check_integer("stride", stride, 0, Rel.GT)
num, channel, height, width = x.shape
out_h = (height + 2*pad - pool_h)//stride + 1
out_w = (width + 2*pad - pool_w)//stride + 1
col = im2col(x, pool_h, pool_w, stride, pad)
col = col.reshape(-1, pool_h*pool_w)
out = np.max(col, axis=1)
out_argmax = np.argmax(col, axis=1)
out = out.reshape(num, out_h, out_w, channel).transpose(0, 3, 1, 2)
out_argmax = out_argmax.reshape(num, out_h, out_w, channel).transpose(0, 3, 1, 2)
return out, out_argmax
def max_pooling(x, pool_h, pool_w, stride):
"""Max pooling."""
validator.check_integer("stride", stride, 0, Rel.GT)
num, channel, height, width = x.shape
out_h = (height - pool_h) // stride + 1
out_w = (width - pool_w) // stride + 1
col = im2col(x, pool_h, pool_w, stride)
col = col.reshape(-1, pool_h * pool_w)
out = np.max(col, axis=1)
out = out.reshape((num, out_h, out_w, channel)).transpose(0, 3, 1, 2)
return out
def conv2d(x, weight, bias=None, stride=1, pad=0,
dilation=1, groups=1, padding_mode='zeros'):
"""Convolution 2D."""
# pylint: disable=unused-argument
validator.check_integer("stride", stride, 0, Rel.GT)
batch_num, _, x_h, x_w = x.shape
filter_num, _, filter_h, filter_w = weight.shape
out_h = 1 + int((x_h + 2 * pad - filter_h - (filter_h - 1) * (dilation - 1)) / stride)
out_w = 1 + int((x_w + 2 * pad - filter_w - (filter_w - 1) * (dilation - 1)) / stride)
col = im2col(x, filter_h, filter_w, stride, pad, dilation)
col_w = np.reshape(weight, (filter_num, -1)).T
out = np.dot(col, col_w)
out = out.reshape(batch_num, out_h, out_w, -1).transpose(0, 3, 1, 2)
if bias is not None:
out += bias
return out
def __init__(self, kernel_size, stride, pad_mode, padding=0, pool=None):
super(_PoolNd, self).__init__()
self.kernel_size = kernel_size
self.stride = stride
self.pad_mode = pad_mode
self.padding = validator.check_integer('padding', padding, 0, Rel.GE)
self.pool = pool
if self.pool is None:
raise NotImplementedError
def _init_depthwise_conv2d(self):
"""Init depthwise conv2d op"""
if context.get_context("device_target") == "Ascend" and self.group > 1:
self.dilation = self._dilation
validator.check_integer('group', self.group, self.in_channels,
Rel.EQ)
validator.check_integer('group', self.group, self.out_channels,
Rel.EQ)
self.conv2d = P.DepthwiseConv2dNative(channel_multiplier=1,
kernel_size=self.kernel_size,
pad_mode=self.pad_mode,
pad=self.padding,
stride=self.stride,
dilation=self.dilation)
weight_shape = [1, self.in_channels, *self.kernel_size]
self.weight = Parameter(initializer(self.weight_init,
weight_shape),
name='weight')
def im2col(img, filter_h, filter_w, stride=1, pad=0, dilation=1):
"""Rearranges an image to row vector."""
validator.check_integer("stride", stride, 0, Rel.GT)
batch_num, channel, height, width = img.shape
out_h = (height + 2*pad - filter_h- (filter_h - 1) * (dilation - 1))//stride + 1
out_w = (width + 2*pad - filter_w- (filter_w - 1) * (dilation - 1))//stride + 1
img = np.pad(img, [(0, 0), (0, 0), (pad, pad), (pad, pad)], 'constant')
col = np.zeros((batch_num, channel, filter_h, filter_w, out_h, out_w)).astype(img.dtype)
for y in range(filter_h):
y_max = y + stride*out_h
for x in range(filter_w):
x_max = x + stride*out_w
col[:, :, y, x, :, :] = img[:, :, y:y_max:stride, x:x_max:stride]
col = col.transpose(0, 4, 5, 1, 2, 3).reshape(batch_num*out_h*out_w, -1)
return col
def col2im(col, input_shape, filter_h, filter_w, stride=1, pad=0):
"""Rearranges a row vector to an image."""
validator.check_integer("stride", stride, 0, Rel.GT)
batch_num, channel, height, width = input_shape
out_h = (height + 2*pad - filter_h)//stride + 1
out_w = (width + 2*pad - filter_w)//stride + 1
col = col.reshape(batch_num, out_h, out_w, channel, filter_h, filter_w) \
.transpose(0, 3, 4, 5, 1, 2)
img = np.zeros((batch_num,
channel,
height + 2*pad + stride - 1,
width + 2*pad + stride - 1)) \
.astype(col.dtype)
for y in range(filter_h):
y_max = y + stride*out_h
for x in range(filter_w):
x_max = x + stride*out_w
img[:, :, y:y_max:stride, x:x_max:stride] += col[:, :, y, x, :, :]
return img[:, :, pad:height + pad, pad:width + pad]
def __init__(self,
in_channels,
out_channels,
kernel_size,
stride=1,
pad_mode='same',
padding=0,
dilation=1,
group=1,
has_bias=False,
weight_init='normal',
bias_init='zeros'):
super(DepthwiseConv2d, self).__init__()
self.kernel_size = twice(kernel_size)
self.stride = twice(stride)
self.dilation = twice(dilation)
self.in_channels = check_int_positive(in_channels)
self.out_channels = check_int_positive(out_channels)
validator.check_integer('group', group, in_channels, Rel.EQ)
validator.check_integer('group', group, out_channels, Rel.EQ)
validator.check_integer('group', group, 1, Rel.GE)
self.pad_mode = pad_mode
self.dilation = dilation
self.group = group
self.has_bias = has_bias
self.weight_init = weight_init
self.bias_init = bias_init
Validator.check_value_type('padding', padding, (int, tuple),
self.cls_name)
if isinstance(padding, tuple):
Validator.check_integer('padding size', len(padding), 4, Rel.EQ,
self.cls_name)
self.padding = padding
self.conv = P.DepthwiseConv2dNative(channel_multiplier=1,
kernel_size=self.kernel_size,
pad_mode=self.pad_mode,
pad=self.padding,
stride=self.stride,
dilation=self.dilation)
self.bias_add = P.BiasAdd()
weight_shape = [1, in_channels, *self.kernel_size]
self.weight = Parameter(initializer(weight_init, weight_shape),
name='weight')
if check_bool(has_bias):
self.bias = Parameter(initializer(bias_init, [out_channels]),
name='bias')
else:
if bias_init != 'zeros':
logger.warning(
"value of `has_bias` is False, value of `bias_init` will be ignore."
)
self.bias = None
def __init__(self,
max_val=1.0,
filter_size=11,
filter_sigma=1.5,
k1=0.01,
k2=0.03):
super(SSIM, self).__init__()
validator.check_type('max_val', max_val, [int, float])
validator.check('max_val', max_val, '', 0.0, Rel.GT)
self.max_val = max_val
self.filter_size = validator.check_integer('filter_size', filter_size,
1, Rel.GE)
self.filter_sigma = validator.check_float_positive(
'filter_sigma', filter_sigma)
validator.check_type('k1', k1, [float])
self.k1 = validator.check_number_range('k1', k1, 0.0, 1.0,
Rel.INC_NEITHER)
validator.check_type('k2', k2, [float])
self.k2 = validator.check_number_range('k2', k2, 0.0, 1.0,
Rel.INC_NEITHER)
self.mean = P.DepthwiseConv2dNative(channel_multiplier=1,
kernel_size=filter_size)