我正在按照here的说明将使用Python生成的Tensorflow模型导入Tensorflow.js。
使用tf.estimator
API创建模型,该模型会根据this page自动保存模型文件。
使用pip安装tensorflowjs
后,运行转换脚本时出现以下错误:
$ tensorflowjs_converter --input_format=tf_saved_model --output_node_names='DeepWater' /home/sterengiannini/deepwater/model-js-import-test /home/sterengiannini/deepwater/web-model
Using TensorFlow backend.
Traceback (most recent call last):
File "/usr/local/bin/tensorflowjs_converter", line 11, in <module>
sys.exit(main())
File "/home/sterengiannini/.local/lib/python2.7/site-packages/tensorflowjs/converters/converter.py", line 317, in main
strip_debug_ops=FLAGS.strip_debug_ops)
File "/home/sterengiannini/.local/lib/python2.7/site-packages/tensorflowjs/converters/tf_saved_model_conversion.py", line 289, in convert_tf_saved_model
input_saved_model_dir=saved_model_dir)
File "/home/sterengiannini/.local/lib/python2.7/site-packages/tensorflow/python/tools/freeze_graph.py", line 338, in freeze_graph
input_saved_model_dir, saved_model_tags).graph_def
File "/home/sterengiannini/.local/lib/python2.7/site-packages/tensorflow/python/tools/saved_model_utils.py", line 42, in get_meta_graph_def
saved_model = reader.read_saved_model(saved_model_dir)
File "/home/sterengiannini/.local/lib/python2.7/site-packages/tensorflow/contrib/saved_model/python/saved_model/reader.py", line 55, in read_saved_model
raise IOError("SavedModel file does not exist at: %s" % saved_model_dir)
IOError: SavedModel file does not exist at: /home/sterengiannini/deepwater/model-js-import-test
以下是模型文件夹中的文件:
$ ls model-js-import-test/
graph.pbtxt model.ckpt-312.data-00000-of-00001 model.ckpt-312.index model.ckpt-312.meta
答案 0 :(得分:1)
该模型似乎没有以正确的格式保存。您应该在模型文件夹(see Structure of a SavedModel directory)中看到以下内容:
assets/
assets.extra/
variables/
variables.data-?????-of-?????
variables.index
saved_model.pb|saved_model.pbtxt
例如,如果您使用的是tf.estimator API,则需要使用以下(see Using SavedModel with Estimators)导出模型:
estimator.export_savedmodel(export_dir_base, serving_input_receiver_fn)