def fc1(X, dropout=True): with tf.name_scope('fc1_layer'): logging.info('FC_1 .........................') X = fc_layers(X, scope_name='fc1') logging.info('FC_1: shape %s', str(X.shape)) # X = batch_norm(X, X.get_shape().as_list()[-1], axis=[0, 1, 2], scope_name='bn2') # logging.info('batch_norm2: shape %s', str(X.shape)) X = activation(X, type='relu') logging.info('RELU_5: shape %s', str(X.shape)) if dropout: X = tf.nn.dropout(X, netParams['fc1']['keep_prob'])#, seed=config.seed_arr[10]) return X
def conv_4(X): with tf.name_scope('conv4_layer'): logging.info('CONV_4 .........................') X = conv_layer(X, scope_name='conv4') logging.info('CONV_4: shape %s', str(X.shape)) X = batch_norm(X, X.get_shape().as_list()[-1], axis=[0, 1, 2], scope_name='bn4') logging.info('BN_3: shape %s', str(X.shape)) X = activation(X, type='relu') logging.info('RELU_4: shape %s', str(X.shape)) X = tf.layers.max_pooling2d(X, pool_size=netParams['conv4']['pool_size'], padding=netParams['conv4']['pool_pad'], strides=netParams['conv4']['pool_stride'], data_format='channels_last') logging.info('MAXPOOL_4: shape %s', str(X.shape)) # X = tf.nn.dropout(X, netParams['conv2']['keep_prob'], seed=config.seed_arr[2]) return X