Unable to Convert Custom EfficientNetB0 Model to Intermediate Representation (IR) Format

• The model was generated by using this code:

  model=tf.keras.applications.EfficientNetB0(

      include_top=True,

      weights=None,

      pooling=max,

      classes=2,

      classifier_activation="softmax"

  )

• Converted the model to SavedModel format
• Ran the Model Optimizer command:

  mo --saved_model_dir model

• Received errors:

  [ ERROR ]  Cannot infer shapes or values for node "StatefulPartitionedCall".

  [ ERROR ]  Error converting shape to a TensorShape: Failed to convert 'masked_array(data=[--, 224, 224, 3],

               mask=[ True, False, False, False],

         fill_value=-1000000007)' to a shape: 'masked'could not be converted to a dimension. A shape should either be single dimension (e.g. 10), or an iterable of dimensions (e.g. [1, 10, None])..

  [ ERROR ] 

  [ ERROR ]  It can happen due to bug in custom shape infer function <function tf_native_tf_node_infer at 0x7ff1ef133310>.

  [ ERROR ]  Or because the node inputs have incorrect values/shapes.

  [ ERROR ]  Or because input shapes are incorrect (embedded to the model or passed via --input_shape).

  [ ERROR ]  Run Model Optimizer with --log_level=DEBUG for more information.

  [ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'openvino.tools.mo.middle.PartialInfer.PartialInfer'>): Stopped shape/value propagation at "StatefulPartitionedCall" node.

   For more information please refer to Model Optimizer FAQ, question #38. (https://docs.openvino.ai/latest/openvino_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html?question=38#question-38) Post Time

The encountered errors are due to some of the layers in the custom model not being compatible with Model Optimizer architecture.

The validated Intel Public Pre-Trained EfficientNet models from the Open Model Zoo are as follows:

• efficientnet-b0
• efficientnet-v2-b0
• efficientnet-v2-s