由於項目需要,需要將TensorFlow保存的模型從ckpt文件轉換為pb文件。
import os
from tensorflow.python import pywrap_tensorflow
from net2use import inception_resnet_v2_small#這里使用自己定義的模型函數即可
import tensorflow as tf
if __name__=='__main__':
pb_file = "./model/output.pb"
ckpt_file = "./model/model.ckpt-652900"
'''
這里的節點名字可能跟設想的有出入,最直接的方法是直接輸出ckpt中保存的節點名字,然后對應着找節點名字,具體的進入convert_variables_to_constants函數的實現中graph_util_impl.py,130行的函數:_assert_nodes_are_present 添加代碼
print('在圖中的節點是:')
for din in name_to_node:
print('{},在圖中'.format(din))
然后運行代碼,若正確就會直接保存;若失敗則會保存失敗,找好輸出節點的名字,在output_node_names 中添加就好
'''
output_node_names = ["embedding"]
with tf.name_scope('input'):
image = tf.placeholder(tf.float32,shape=(None,79,199,1),name='input_image')
net, endpoints=inception_resnet_v2_small(image, is_training=False)
embedding = tf.nn.l2_normalize(net,1,1e-10,name='embedding')
config=tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.per_process_gpu_memory_fraction = 0.45
sess = tf.Session(config = config)
saver = tf.train.Saver()
saver.restore(sess, ckpt_file)
print('read success')
converted_graph_def = tf.graph_util.convert_variables_to_constants(sess,
input_graph_def = sess.graph.as_graph_def(),
output_node_names = output_node_names)
with tf.gfile.GFile(pb_file, "wb") as f:
f.write(converted_graph_def.SerializeToString())
print('保存成功')