# ==============================================================================
# Copyright (C) 2020 Vladimir Juras, Ravinder Regatte and Cem M. Deniz
#
# This file is part of 2019_IWOAI_Challenge
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
# ==============================================================================
import tensorflow as tf
def _variable_on_cpu(name, shape, initializer):
'''Helper to create a Variable stored on CPU memory.
Args:
name: name of the variable
shape: list of ints
initializer: initializer for Variable
Returns:
Variable Tensor
'''
with tf.device('/cpu:0'):
var = tf.get_variable(name, shape, initializer=initializer)
return var
def _variable_with_weight_decay(name, shape, scale, initializer=tf.contrib.layers.variance_scaling_initializer()):
#Helper to create an initialized Variable with weight decay.
with tf.device('/cpu:0'):
var = tf.get_variable(name, shape, initializer=initializer,
regularizer=tf.contrib.layers.l2_regularizer(scale))
return var
def conv3d(name, bottom, num_output, kernel_size=[3,3,3], reg_constant=0.0, strides=[1,1,1,1,1],
padding='SAME', initializer=tf.contrib.layers.variance_scaling_initializer(), bias=True):
bottom_shape = bottom.get_shape()
with tf.variable_scope(name) as scope:
kernel = _variable_with_weight_decay('weights', shape=[kernel_size[0], kernel_size[1], kernel_size[2], bottom_shape[4], num_output],
scale=reg_constant, initializer=initializer)
conv = tf.nn.conv3d(bottom, kernel, strides, padding=padding)
if bias:
biases = _variable_on_cpu('biases', [num_output], initializer=tf.constant_initializer(0.0))
top = tf.nn.bias_add(conv, biases, name=scope.name)
else:
top = conv
print (name,padding, top.get_shape())
#tf.summary.image(name, _get_image_summary(top), max_outputs=4)
return top
def downconv3d(name, bottom, num_output, kernel_size=[2,2,2], reg_constant=0.0, strides=[1,2,2,2,1],
padding='SAME', initializer=tf.contrib.layers.variance_scaling_initializer()):
bottom_shape = bottom.get_shape()
with tf.variable_scope(name) as scope:
kernel = _variable_with_weight_decay('weights', shape=[kernel_size[0], kernel_size[1], kernel_size[2], bottom_shape[4], num_output],
scale=reg_constant, initializer=initializer)
conv = tf.nn.conv3d(bottom, kernel, strides, padding=padding)
biases = _variable_on_cpu('biases', [num_output], initializer=tf.constant_initializer(0.0))
top = tf.nn.bias_add(conv, biases, name=scope.name)
print (name, top.get_shape())
#tf.summary.image(name, _get_image_summary(top))
return top
def max_pool(name, bottom):
with tf.variable_scope(name) as scope:
top = tf.nn.max_pool3d(input=bottom, ksize=[1,2,2,2,1], strides=[1,2,2,2,1], padding='SAME', name=name)
print (name, top.get_shape())
#tf.summary.image(name, _get_image_summary(top))
return top
def upconv3d(name, bottom, num_output, out_value_shape, kernel_size=[2,2,2], reg_constant=0.0, strides=[1, 2, 2, 2, 1],
padding='SAME', initializer=tf.contrib.layers.variance_scaling_initializer()):
batch_size = tf.shape(bottom)[0]
bottom_shape = bottom.get_shape()
output_shape = tf.stack([batch_size, out_value_shape[0], out_value_shape[1], out_value_shape[2], num_output])
with tf.variable_scope(name) as scope:
kernel = _variable_with_weight_decay('weights', shape=[kernel_size[0], kernel_size[1], kernel_size[2], num_output, bottom_shape[4]],
scale=reg_constant, initializer=initializer)
conv = tf.nn.conv3d_transpose(bottom, kernel, output_shape, strides, padding=padding)
biases = _variable_on_cpu('biases', [num_output], initializer=tf.constant_initializer(0.0))
top = tf.nn.bias_add(conv, biases, name=scope.name)
print (name, top.get_shape())
#tf.summary.image(name, _get_image_summary(top))
return top
def atrousconv3d(name, bottom, num_output, kernel_size=[3,3,3], reg_constant=0.0, strides=[1, 1, 1], dilation_rate=[2,2,1],
padding='SAME', initializer=tf.contrib.layers.variance_scaling_initializer()):
batch_size = tf.shape(bottom)[0]
bottom_shape = bottom.get_shape()
with tf.variable_scope(name) as scope:
kernel = _variable_with_weight_decay('weights', shape=[kernel_size[0], kernel_size[1], kernel_size[2], num_output, bottom_shape[4]],
scale=reg_constant, initializer=initializer)
conv = tf.nn.convolution(bottom, kernel, padding=padding, strides=strides, dilation_rate=dilation_rate)
biases = _variable_on_cpu('biases', [num_output], initializer=tf.constant_initializer(0.0))
top = tf.nn.bias_add(conv, biases, name=scope.name)
print (name, top.get_shape())
#tf.summary.image(name, _get_image_summary(top))
return top
def relu(name, bottom):
with tf.variable_scope(name) as scope:
top = tf.nn.relu(bottom, name=name)
#tf.summary.histogram(name + '/activations', top)
#tf.summary.image(name, _get_image_summary(top), max_outputs=4)
return top
def add_res(name, bottom, res, conv=True, reg_constant=0.0, skip=False):
if skip:
return bottom
if conv:
bottom_shape = bottom.get_shape()
res_shape = res.get_shape()
with tf.variable_scope(name) as scope:
kernel = _variable_with_weight_decay('weights', shape=[1, 1, 1, res_shape[4], bottom_shape[4]],
scale=reg_constant)
conv = tf.nn.conv3d(res, kernel, strides=[1,1,1,1,1], padding='SAME')
top = tf.add(bottom, conv, name=name)
else:
with tf.variable_scope(name) as scope:
top = tf.add(bottom, res, name=name)
#tf.summary.image(name, _get_image_summary(top))
return top
def concat(name, x1, x2):
with tf.variable_scope(name) as scope:
top = tf.concat([x1, x2], 4, name=name)
print (name, top.get_shape())
#tf.summary.image(name, _get_image_summary(top))
return top
def multiconcat(name, xs):
with tf.variable_scope(name) as scope:
top = tf.concat(xs, 4, name=name)
print (name, top.get_shape())
#tf.summary.image(name, _get_image_summary(top))
return top
def dropout(name, bottom, keep_prob=0.5):
top = tf.nn.dropout(bottom, keep_prob, name=name)
return top
def _get_image_summary(img, idx=0):
'''
Make an image summary for 5d tensor image with index idx
'''
img_z = tf.shape(img)[3]
V = tf.slice(img, (0, 0, 0, img_z//2-1, idx), (1, -1, -1, 3, 1))
V -= tf.reduce_min(V)
V /= tf.reduce_max(V)
V *= 255
img_w = tf.shape(img)[1]
img_h = tf.shape(img)[2]
V = tf.reshape(V, tf.stack((img_w, img_h, -1)))
V = tf.transpose(V, (2, 0, 1))
V = tf.reshape(V, tf.stack((-1, img_w, img_h, 1)))
return V
Datasets
Models
