resnet Namespace Reference

Classes
class	ResnetBuilder

Functions
def	_bn_relu (input)
def	_conv_bn_relu (conv_params)
def	_bn_relu_conv (conv_params)
def	_shortcut (input, residual)
def	_residual_block (block_function, filters, repetitions, is_first_layer=False)
def	basic_block (filters, init_strides=(1, 1), is_first_block_of_first_layer=False)
def	bottleneck (filters, init_strides=(1, 1), is_first_block_of_first_layer=False)
def	_handle_dim_ordering ()
def	_get_block (identifier)

Detailed Description

Based on https://github.com/raghakot/keras-resnet/blob/master/resnet.py

Function Documentation

def resnet._bn_relu (   input )

private

Helper to build a BN -> relu block

Definition at line 33 of file resnet.py.

def _bn_relu(input):
    """Helper to build a BN -> relu block
    """
    norm = BatchNormalization(axis=CHANNEL_AXIS)(input)
    return Activation("relu")(norm)

def resnet._bn_relu_conv (   conv_params )

private

Helper to build a BN -> relu -> conv block.
This is an improved scheme proposed in http://arxiv.org/pdf/1603.05027v2.pdf

Definition at line 60 of file resnet.py.

def _bn_relu_conv(**conv_params):
    """Helper to build a BN -> relu -> conv block.
    This is an improved scheme proposed in http://arxiv.org/pdf/1603.05027v2.pdf
    """
    filters = conv_params["filters"]
    kernel_size = conv_params["kernel_size"]
    strides = conv_params.setdefault("strides", (1, 1))
    kernel_initializer = conv_params.setdefault("kernel_initializer", "he_normal")
    padding = conv_params.setdefault("padding", "same")
    kernel_regularizer = conv_params.setdefault("kernel_regularizer", l2(1.e-4))

    def f(input):
        activation = _bn_relu(input)
        return Conv2D(filters=filters, kernel_size=kernel_size,
                      strides=strides, padding=padding,
                      kernel_initializer=kernel_initializer,
                      kernel_regularizer=kernel_regularizer)(activation)

    return f

def resnet._conv_bn_relu (   conv_params )

private

Helper to build a conv -> BN -> relu block

Definition at line 40 of file resnet.py.

def _conv_bn_relu(**conv_params):
    """Helper to build a conv -> BN -> relu block
    """
    filters = conv_params["filters"]
    kernel_size = conv_params["kernel_size"]
    strides = conv_params.setdefault("strides", (1, 1))
    kernel_initializer = conv_params.setdefault("kernel_initializer", "he_normal")
    padding = conv_params.setdefault("padding", "same")
    kernel_regularizer = conv_params.setdefault("kernel_regularizer", l2(1.e-4))

    def f(input):
        conv = Conv2D(filters=filters, kernel_size=kernel_size,
                      strides=strides, padding=padding,
                      kernel_initializer=kernel_initializer,
                      kernel_regularizer=kernel_regularizer)(input)
        return _bn_relu(conv)

    return f

def resnet._get_block (   identifier )

private

Definition at line 185 of file resnet.py.

def _get_block(identifier):
    if isinstance(identifier, six.string_types):
        res = globals().get(identifier)
        if not res:
            raise ValueError('Invalid {}'.format(identifier))
        return res
    return identifier

def resnet._handle_dim_ordering ( )

private

Definition at line 171 of file resnet.py.

def _handle_dim_ordering():
    global ROW_AXIS
    global COL_AXIS
    global CHANNEL_AXIS
    if K.image_dim_ordering() == 'tf':
        ROW_AXIS = 1
        COL_AXIS = 2
        CHANNEL_AXIS = 3
    else:
        CHANNEL_AXIS = 1
        ROW_AXIS = 2
        COL_AXIS = 3

def resnet._residual_block (   block_function,   filters,   repetitions,   is_first_layer = False  )

private

Builds a residual block with repeating bottleneck blocks.

Definition at line 106 of file resnet.py.

def _residual_block(block_function, filters, repetitions, is_first_layer=False):
    """Builds a residual block with repeating bottleneck blocks.
    """
    def f(input):
        for i in range(repetitions):
            init_strides = (1, 1)
            if i == 0 and not is_first_layer:
                init_strides = (2, 2)
            input = block_function(filters=filters, init_strides=init_strides,
                                   is_first_block_of_first_layer=(is_first_layer and i == 0))(input)
        return input

    return f

def resnet._shortcut (   input,   residual  )

private

Adds a shortcut between input and residual block and merges them with "sum"

Definition at line 81 of file resnet.py.

def _shortcut(input, residual):
    """Adds a shortcut between input and residual block and merges them with "sum"
    """
    # Expand channels of shortcut to match residual.
    # Stride appropriately to match residual (width, height)
    # Should be int if network architecture is correctly configured.
    input_shape = K.int_shape(input)
    residual_shape = K.int_shape(residual)
    stride_width = int(round(input_shape[ROW_AXIS] / residual_shape[ROW_AXIS]))
    stride_height = int(round(input_shape[COL_AXIS] / residual_shape[COL_AXIS]))
    equal_channels = input_shape[CHANNEL_AXIS] == residual_shape[CHANNEL_AXIS]

    shortcut = input
    # 1 X 1 conv if shape is different. Else identity.
    if stride_width > 1 or stride_height > 1 or not equal_channels:
        shortcut = Conv2D(filters=residual_shape[CHANNEL_AXIS],
                          kernel_size=(1, 1),
                          strides=(stride_width, stride_height),
                          padding="valid",
                          kernel_initializer="he_normal",
                          kernel_regularizer=l2(0.0001))(input)

    return add([shortcut, residual])

def resnet.basic_block	(		filters,
			init_strides = (1, 1),
			is_first_block_of_first_layer = False
	)

Basic 3 X 3 convolution blocks for use on resnets with layers <= 34.
Follows improved proposed scheme in http://arxiv.org/pdf/1603.05027v2.pdf

Definition at line 121 of file resnet.py.

def basic_block(filters, init_strides=(1, 1), is_first_block_of_first_layer=False):
    """Basic 3 X 3 convolution blocks for use on resnets with layers <= 34.
    Follows improved proposed scheme in http://arxiv.org/pdf/1603.05027v2.pdf
    """
    def f(input):

        if is_first_block_of_first_layer:
            # don't repeat bn->relu since we just did bn->relu->maxpool
            conv1 = Conv2D(filters=filters, kernel_size=(3, 3),
                           strides=init_strides,
                           padding="same",
                           kernel_initializer="he_normal",
                           kernel_regularizer=l2(1e-4))(input)
        else:
            conv1 = _bn_relu_conv(filters=filters, kernel_size=(3, 3),
                                  strides=init_strides)(input)

        residual = _bn_relu_conv(filters=filters, kernel_size=(3, 3))(conv1)
        return _shortcut(input, residual)

    return f

def resnet.bottleneck	(		filters,
			init_strides = (1, 1),
			is_first_block_of_first_layer = False
	)

Bottleneck architecture for > 34 layer resnet.
Follows improved proposed scheme in http://arxiv.org/pdf/1603.05027v2.pdf

Returns:
    A final conv layer of filters * 4

Definition at line 144 of file resnet.py.

def bottleneck(filters, init_strides=(1, 1), is_first_block_of_first_layer=False):
    """Bottleneck architecture for > 34 layer resnet.
    Follows improved proposed scheme in http://arxiv.org/pdf/1603.05027v2.pdf

    Returns:
        A final conv layer of filters * 4
    """
    def f(input):

        if is_first_block_of_first_layer:
            # don't repeat bn->relu since we just did bn->relu->maxpool
            conv_1_1 = Conv2D(filters=filters, kernel_size=(1, 1),
                              strides=init_strides,
                              padding="same",
                              kernel_initializer="he_normal",
                              kernel_regularizer=l2(1e-4))(input)
        else:
            conv_1_1 = _bn_relu_conv(filters=filters, kernel_size=(1, 1),
                                     strides=init_strides)(input)

        conv_3_3 = _bn_relu_conv(filters=filters, kernel_size=(3, 3))(conv_1_1)
        residual = _bn_relu_conv(filters=filters * 4, kernel_size=(1, 1))(conv_3_3)
        return _shortcut(input, residual)

    return f