使用tflite进行训练后量化会导致运行时错误

Question

我正在尝试量化我的模型(特别是从模型动物园下载的关于coco的预先训练的faster_rcnn_inception_v2 )，希望加速推理时间。

我使用here中的以下代码

import tensorflow as tf
converter = tf.lite.TocoConverter.from_saved_model(saved_model_dir)
converter.post_training_quantize = True
tflite_quantized_model = converter.convert()
open("quantized_model.tflite", "wb").write(tflite_quantized_model)

模型目录没有saved_model.pb文件。因此，我将frozen_inference_graph.pb重命名为saved_model.pb。

运行上面的代码会产生以下运行时错误：

Traceback (most recent call last):
  File "/home/juggernaut/pycharm-community-2018.2.4/helpers/pydev/pydevd.py", line 1664, in <module>
    main()
  File "/home/juggernaut/pycharm-community-2018.2.4/helpers/pydev/pydevd.py", line 1658, in main
    globals = debugger.run(setup['file'], None, None, is_module)
  File "/home/juggernaut/pycharm-community-2018.2.4/helpers/pydev/pydevd.py", line 1068, in run
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "/hdd/motorola/motorola_heads/tensorflow_face_detection/quantize.py", line 5, in <module>
    converter = tf.lite.TocoConverter.from_saved_model(saved_model_dir)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/python/util/deprecation.py", line 318, in new_func
    return func(*args, **kwargs)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/lite/python/lite.py", line 587, in from_saved_model
    tag_set, signature_key)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/lite/python/lite.py", line 376, in from_saved_model
    output_arrays, tag_set, signature_key)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/lite/python/convert_saved_model.py", line 254, in freeze_saved_model
    meta_graph = get_meta_graph_def(saved_model_dir, tag_set)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/lite/python/convert_saved_model.py", line 61, in get_meta_graph_def
    return loader.load(sess, tag_set, saved_model_dir)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/python/util/deprecation.py", line 318, in new_func
    return func(*args, **kwargs)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/python/saved_model/loader_impl.py", line 269, in load
    return loader.load(sess, tags, import_scope, **saver_kwargs)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/python/saved_model/loader_impl.py", line 420, in load
    **saver_kwargs)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/python/saved_model/loader_impl.py", line 347, in load_graph
    meta_graph_def = self.get_meta_graph_def_from_tags(tags)
  File "/hdd/motorola/venv_py27_tf1.10/local/lib/python2.7/site-packages/tensorflow/python/saved_model/loader_impl.py", line 323, in get_meta_graph_def_from_tags
    " could not be found in SavedModel. To inspect available tag-sets in"
RuntimeError: MetaGraphDef associated with tags set(['serve']) could not be found in SavedModel. To inspect available tag-sets in the SavedModel, please use the SavedModel CLI: `saved_model_cli`

这意味着什么，我应该怎么做？

Answer 1

请参考this issue。他们似乎和你有同样的问题。

在较新版本的Tensorflow中可能会修复此问题(可能在此期间，标签已从'serving‘切换为’serving‘)。

Answer 2

您应该使用tf.saved_model.simple_save来保存pb模型。