def build_drn_block(inputdrn, dim, name="drn", padding="REFLECT", norm_type=None, is_training=True, keep_rate=0.75):
with tf.variable_scope(name):
out_drn = tf.pad(inputdrn, [[0, 0], [2, 2], [2, 2], [0, 0]], padding)
out_drn = layers.dilate_conv2d(out_drn, dim, dim, 3, 3, 2, 0.01, "VALID", "c1", norm_type=norm_type, is_training=is_training, keep_rate=keep_rate)
out_drn = tf.pad(out_drn, [[0, 0], [2, 2], [2, 2], [0, 0]], padding)
out_drn = layers.dilate_conv2d(out_drn, dim, dim, 3, 3, 2, 0.01, "VALID", "c2", do_relu=False, norm_type=norm_type, is_training=is_training, keep_rate=keep_rate)
return tf.nn.relu(out_drn + inputdrn)
def build_drn_block_ds(inputdrn, dim_in, dim_out, name='drn_ds', padding="REFLECT", norm_type=None, is_training=True, keep_rate=0.75):
with tf.variable_scope(name):
out_drn = tf.pad(inputdrn, [[0,0], [2,2], [2,2], [0,0]], padding)
out_drn = layers.dilate_conv2d(out_drn, dim_in, dim_out, 3, 3, 2, 0.01, 'VALID', "c1", norm_type=norm_type, is_training=is_training, keep_rate=keep_rate)
out_drn = tf.pad(out_drn, [[0,0], [2,2], [2,2], [0,0]], padding)
out_drn = layers.dilate_conv2d(out_drn, dim_out, dim_out, 3, 3, 2, 0.01, 'VALID', "c2", do_relu=False, norm_type=norm_type, is_training=is_training, keep_rate=keep_rate)
inputdrn = tf.pad(inputdrn, [[0,0], [0,0], [0, 0], [(dim_out-dim_in)//2,(dim_out-dim_in)//2]], padding)
return tf.nn.relu(out_drn + inputdrn)