def reshape_temporal_conv1d_bn(inputs,
is_training,
filters,
kernel_size,
num_frames=32,
stride=1,
use_relu=True,
data_format='channels_last'):
"""Performs 1D temporal conv.
followed by batch normalization with reshaping.
Args:
inputs: `Tensor` of size `[batch*time, height, width, channels]`. Only
supports 'channels_last' as the data format.
is_training: 'bool' specifying whether in training mode or not.
filters: `int` number of filters in the convolution.
kernel_size: `int` kernel size to be used for `conv2d` or max_pool2d`
operations. Should be a positive integer.
num_frames: 'int' number of frames in the input tensor.
stride: 'int' temporal stride
use_relu: 'boolean' determines if the activation function is used or not.
data_format: `str`. Only supports 'channels_last' as the data format.
Returns:
A padded `Tensor` of the same `data_format` with size either intact
(if `kernel_size == 1`) or padded (if `kernel_size > 1`).
"""
assert data_format == 'channels_last'
feature_shape = inputs.shape
batch_size = feature_shape[0] // num_frames
inputs = tf.reshape(inputs, [
int(feature_shape[0] // num_frames), num_frames,
feature_shape[1] * feature_shape[2], -1
])
inputs = tf.layers.conv2d(
inputs=inputs,
filters=filters,
kernel_size=(kernel_size, 1),
strides=[stride, 1],
padding='SAME',
use_bias=False,
kernel_initializer=contrib_layers.variance_scaling_initializer(
factor=2.0, mode='FAN_IN', uniform=False),
data_format=data_format)
new_frames = inputs.shape[1]
inputs = tf.reshape(
inputs, [batch_size * new_frames, feature_shape[1], feature_shape[2], -1])
inputs = bn.batch_norm_relu(
inputs,
is_training,
relu=use_relu,
data_format=data_format)
return inputs, new_frames
def conv2d(inputs,
kernel,
filters,
stride,
is_training,
use_relu=True,
init_zero=False,
use_bn=True,
data_format='channels_last'):
"""Performs 2d conv.
Args:
inputs: `Tensor` of size `[batch*time, height, width, channels]`. Only
supports 'channels_last' as the data format.
kernel: `int` kernel size to be used for `conv2d` or max_pool2d` operations.
Should be a positive integer.
filters: `int` number of filters in the convolution.
stride: 'int' temporal stride
is_training: 'bool' specifying whether in training mode or not.
use_relu: bool, apply relu after batch norm
init_zero: bool, initialize batch_norm weights to 0 (i.e., initially ignore
layer)
use_bn: bool, use batch norm or not
data_format: `str`. Only supports 'channels_last' as the data format.
Returns:
A `Tensor` of the same data_format
"""
inputs = tf.layers.conv2d(
inputs=inputs,
filters=filters,
kernel_size=kernel,
strides=stride,
padding='SAME',
use_bias=False,
kernel_initializer=contrib_layers.variance_scaling_initializer(
factor=2.0, mode='FAN_IN', uniform=False),
data_format=data_format)
if use_bn:
inputs = bn.batch_norm_relu(
inputs,
is_training,
relu=use_relu,
init_zero=init_zero,
data_format=data_format)
if not use_bn and use_relu:
inputs = tf.nn.relu(inputs)
return inputs
def spatial_conv(inputs,
conv_type,
kernel,
filters,
stride,
is_training,
activation_fn='relu',
data_format='channels_last'):
"""Performs 1x1 conv followed by 2d or depthwise conv.
Args:
inputs: `Tensor` of size `[batch*time, height, width, channels]`. Only
supports 'channels_last' as the data format.
conv_type: 'string' of "std", "depth", "maxpool", or "avgpool" this selects
the spatial conv/pooling method.
kernel: `int` kernel size to be used for `conv2d` or max_pool2d` operations.
Should be a positive integer.
filters: `int` number of filters in the convolution.
stride: 'int' temporal stride
is_training: 'bool' specifying whether in training mode or not.
activation_fn: 'string' the activation function to use (relu or swish)
data_format: `str`. Only supports 'channels_last' as the data format.
Returns:
A `Tensor` of the same data_format
"""
if kernel == 1:
return inputs
use_relu = (activation_fn == 'relu')
if conv_type == 'std' or conv_type == 'depth':
inputs = conv2d(inputs, 1, filters, 1, is_training, use_relu=use_relu)
if not use_relu:
inputs = hard_swish(inputs)
if conv_type == 'std' or conv_type == '1std':
inputs = conv2d(inputs, int(kernel), filters, int(stride), is_training,
use_relu=use_relu)
if not use_relu:
inputs = hard_swish(inputs)
elif conv_type == 'depth':
depth_multiplier = 1
depthwise_kernel_shape = (int(kernel), int(kernel), inputs.shape[-1],
depth_multiplier)
depthwise_kernel = contrib_framework.model_variable(
name='depthwise_kernel',
shape=depthwise_kernel_shape,
dtype=tf.float32,
initializer=contrib_layers.variance_scaling_initializer(
factor=2.0, mode='FAN_IN', uniform=False),
trainable=True)
inputs = tf.nn.depthwise_conv2d(
inputs,
tf.cast(depthwise_kernel, inputs.dtype),
strides=[1, int(stride), int(stride), 1],
padding='SAME',
rate=[1, 1],
data_format='NHWC' if data_format == 'channels_last' else 'NCHW')
inputs = bn.batch_norm_relu(
inputs,
is_training,
relu=use_relu,
data_format=data_format)
if not use_relu:
inputs = hard_swish(inputs)
elif conv_type == 'maxpool':
inputs = tf.layers.max_pooling2d(
inputs,
int(kernel),
int(stride),
padding='same',
data_format=data_format)
elif conv_type == 'avgpool':
inputs = tf.layers.average_pooling2d(
inputs,
int(kernel),
int(stride),
padding='same',
data_format=data_format)
return inputs