Tensorflow 2对象检测API-将. ckpt文件转换为. pb/任何保存的模型格式，用于简单的Web应用部署

Question

我正在尝试使用TensorFlow model zoo的ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8模型使用TensorFlow对象检测API进行对象检测。我能够使用ckpt文件检测单个测试图像（保存为ckpt-17.data-00000-of-00001）

我需要将此ckpt转换为一些保存的模型文件（.pb/.h5），以便在简单的flask webapp中使用。

我发现很难将ckpt文件转换为。pb使用下面的代码

print("""python {}/research/object_detection/export_inference_graph.py \
--input_type image_tensor \
--pipeline_config_path={}/{}/pipeline.config \
--trained /content/RealTimeObjectDetectionStages/Tensorflow/workspace/models/my_ssd_mobnet/ckpt-17.index \
--output_directory /content/RealTimeObjectDetectionStages/Tensorflow/workspace/models""".format(APIMODEL_PATH, MODEL_PATH,CUSTOM_MODEL_NAME,MODEL_PATH,CUSTOM_MODEL_NAME))

!python /content/RealTimeObjectDetectionStages/Tensorflow/models/research/object_detection/export_inference_graph.py --input_type image_tensor --pipeline_config_path=/content/RealTimeObjectDetectionStages/Tensorflow/workspace/models/my_ssd_mobnet/pipeline.config --trained /content/RealTimeObjectDetectionStages/Tensorflow/workspace/models/my_ssd_mobnet/ckpt-17.ckpt --output_directory /content/RealTimeObjectDetectionStages/Tensorflow/workspace/models

当我试图转换为时，会出现以下错误。PB

2021-03-12 11:25:57.212865: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/absl/flags/_flagvalues.py", line 541, in _assert_validators
    validator.verify(self)
  File "/usr/local/lib/python3.7/dist-packages/absl/flags/_validators.py", line 82, in verify
    raise _exceptions.ValidationError(self.message)
absl.flags._exceptions.ValidationError: Flag --trained_checkpoint_prefix must have a value other than None.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/content/RealTimeObjectDetectionStages/Tensorflow/models/research/object_detection/export_inference_graph.py", line 206, in 
    tf.app.run()
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/platform/app.py", line 40, in run
    _run(main=main, argv=argv, flags_parser=_parse_flags_tolerate_undef)
  File "/usr/local/lib/python3.7/dist-packages/absl/app.py", line 294, in run
    flags_parser,
  File "/usr/local/lib/python3.7/dist-packages/absl/app.py", line 363, in _run_init
    flags_parser=flags_parser,
  File "/usr/local/lib/python3.7/dist-packages/absl/app.py", line 213, in _register_and_parse_flags_with_usage
    args_to_main = flags_parser(original_argv)
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/platform/app.py", line 31, in _parse_flags_tolerate_undef
    return flags.FLAGS(_sys.argv if argv is None else argv, known_only=True)
  File "/usr/local/lib/python3.7/dist-packages/tensorflow/python/platform/flags.py", line 113, in __call__
    return self.__dict__['__wrapped'].__call__(*args, **kwargs)
  File "/usr/local/lib/python3.7/dist-packages/absl/flags/_flagvalues.py", line 649, in __call__
    self.validate_all_flags()
  File "/usr/local/lib/python3.7/dist-packages/absl/flags/_flagvalues.py", line 523, in validate_all_flags
    self._assert_validators(all_validators)
  File "/usr/local/lib/python3.7/dist-packages/absl/flags/_flagvalues.py", line 544, in _assert_validators
    raise _exceptions.IllegalFlagValueError('%s: %s' % (message, str(e)))
absl.flags._exceptions.IllegalFlagValueError: flag --trained_checkpoint_prefix=None: Flag --trained_checkpoint_prefix must have a value other than None.

如何保存检查点以供部署？

我正在使用谷歌colab的TensorFlow 2

http://stackoverflow.com/questions/tagged/tensorflow-model-garden

Answer 1

更正文件路径格式解决了问题：

!python /content/RealTimeObjectDetectionStages/Tensorflow/models/research/object_detection/model_main_tf2.py --model_dir=/content/RealTimeObjectDetectionStages/Tensorflow/workspace/models/my_ssd_mobnet --pipeline_config_path=/content/RealTimeObjectDetectionStages/Tensorflow/workspace/models/my_ssd_mobnet/pipeline.config --num_train_steps=1000