xception65をkeras(h5)に変換したい

[tatusya-suda]

Issue Type

Others

OS

Windows

OS architecture

x86_64

Programming Language

Python

Framework

TensorFlow

Download URL for ONNX / OpenVINO IR

https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md

Convert Script

!source /opt/intel/openvino_2021/bin/setupvars.sh && pb_to_saved_model \ --pb_file_path frozen_inference_graph.pb \ --inputs sub_2:0 \ --outputs ResizeBilinear_2:0

!source /opt/intel/openvino_2021/bin/setupvars.sh &&\ python3 /opt/intel/openvino_2021/deployment_tools/model_optimizer/mo.py \ --saved_model_dir saved_model_from_pb/ --output_dir model_saved --data_type FP32

Description

https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md のxception65_coco_voc_trainaug(.pb)をkeras(h5)に変換したいです。 しかし、openvino2tensorflowを使用して変換するときにpickleエラーが起きます。

環境はpython=3.8,tensorflow=2.5です。

変換ルートはpbからsaved_modelに変換したあと、IR形式に変換しています。

Relevant Log Output

TypeError: can't pickle module objects

Source code for simple inference testing code

No response

[PINTO0309]

1. Please let me know why you want to convert to .h5. This is a deprecated format at this time. saved_model should be sufficient for your needs.
2. Is this the model you are converting to? deeplabv3_pascal_train_aug_2018_01_04.tar.gz
3. Your specification of the name of the input/output layer in your .pb is incorrect.

   pb_to_saved_model \
   --pb_file_path frozen_inference_graph.pb \
   --inputs ImageTensor:0 \
   --outputs SemanticPredictions:0

   

saved_model_to_tflite \
--saved_model_dir_path saved_model_from_pb \
--input_shape [1,513,513,3] \
--model_output_dir_path saved_model \
--output_no_quant_float32_tflite

cp saved_model/model_float32.tflite .

tflite2tensorflow \
--model_path model_float32.tflite \
--flatc_path ../flatc \
--schema_path ../schema.fbs \
--output_pb \
--optimizing_for_openvino_and_myriad

saved_model_cli show --dir saved_model --all

2022-08-28 08:57:19.305342: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  SSE3 SSE4.1 SSE4.2 AVX AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['ImageTensor'] tensor_info:
        dtype: DT_UINT8
        shape: (1, 513, 513, 3)
        name: ImageTensor:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['SemanticPredictions'] tensor_info:
        dtype: DT_INT64
        shape: (1, 513, 513)
        name: SemanticPredictions:0
  Method name is: tensorflow/serving/predict

[tatusya-suda]

1.keras modelに変換して、モデルの剪定を実施したいです。 剪定したモデルをtfliteに変換し、Androidへの実装を目指しています。 https://www.tensorflow.org/model_optimization/guide/pruning/pruning_with_keras

2.モデルはご指定のモデルです。

3.調べたときsub2とResizeBilinearを指定した例があったのでそちらで変換していました。ありがとうございます

[PINTO0309]

If that is the case, then this question is not about openvino2tensorflow, but about how to use TensorFlow.

First, delete all unnecessary garbage processing in the range shown below.

• export_model.py https://github.com/tensorflow/models/blob/1587d2db03455264fdc09b246a1ee56b88616499/research/deeplab/export_model.py#L91-L111
• model structure
• Reference TensorFlow2でのモデルファイルの保存と読み込み