Trouble converting vehicle-detection-0200

[hesamsheikh]

I'm converting the openvino vehicle-detection-0200 model. Though the last DetectionOutput layer is not implemented. I changed the xml to this:

I removed the last two layers of the original model and didn't change the .bin weight file. I give the new xml to openvino2tensorflow.py and it runs with no errors. Though the model it gives me only has the input layer and is less than 100kb. Did I do something wrong? cp.txt

[PINTO0309]

Perhaps you are editing it wrong. I can't read anything that looks like the xml file you posted, but I guess you just haven't reconnected the Result layer.

[hesamsheikh]

I replaced the xml text with txt file you can view.

[PINTO0309]

There is no Result layer.

[PINTO0309]

vehicle-detection-0200.xml.zip

[PINTO0309]

https://github.com/PINTO0309/PINTO_model_zoo/tree/main/178_vehicle-detection-0200

[PINTO0309]

https://github.com/PINTO0309/openvino2tensorflow/releases/tag/v1.25.3 https://github.com/PINTO0309/openvino2tensorflow/releases/tag/v1.25.4

[hesamsheikh]

https://github.com/PINTO0309/PINTO_model_zoo/tree/main/178_vehicle-detection-0200

Thanks for the effort. Though it seems that the saved_model format you have provided has some problems. It's in the ''_UserObject' object' format and normal functions like predict and summary cannot be applied to it.

[PINTO0309]

>>> import tensorflow as tf
>>> import numpy as np
>>> loaded = tf.saved_model.load('saved_model_256x256')
>>> infer = loaded.signatures["serving_default"]
>>> infer(image=np.ones((1,256,256,3),dtype=np.float32))
2021-11-20 21:36:29.388833: I tensorflow/stream_executor/cuda/cuda_dnn.cc:369] Loaded cuDNN version 8200
{'tf.identity': <tf.Tensor: shape=(1, 5376), dtype=float32, numpy=
array([[-0.6882432 , -0.92401135,  0.3261156 , ..., -0.20971192,
        -0.18629664, -0.2545644 ]], dtype=float32)>, 'tf.identity_1': <tf.Tensor: shape=(1, 2688), dtype=float32, numpy=
array([[3.9097137e-04, 9.9960905e-01, 8.8063441e-04, ..., 9.9621165e-01,
        3.4444174e-03, 9.9655557e-01]], dtype=float32)>}

$ saved_model_cli show --dir saved_model_256x256/ --all

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

signature_def['__saved_model_init_op']:
  The given SavedModel SignatureDef contains the following input(s):
  The given SavedModel SignatureDef contains the following output(s):
    outputs['__saved_model_init_op'] tensor_info:
        dtype: DT_INVALID
        shape: unknown_rank
        name: NoOp
  Method name is: 

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['image'] tensor_info:
        dtype: DT_FLOAT
        shape: (1, 256, 256, 3)
        name: serving_default_image:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['tf.identity'] tensor_info:
        dtype: DT_FLOAT
        shape: (1, 5376)
        name: StatefulPartitionedCall:0
    outputs['tf.identity_1'] tensor_info:
        dtype: DT_FLOAT
        shape: (1, 2688)
        name: StatefulPartitionedCall:1
  Method name is: tensorflow/serving/predict

Defined Functions:
  Function Name: '__call__'
    Option #1
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None
    Option #2
      Callable with:
        Argument #1
          image: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='image')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None
    Option #3
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None
    Option #4
      Callable with:
        Argument #1
          image: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='image')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None

  Function Name: '_default_save_signature'
    Option #1
      Callable with:
        Argument #1
          image: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='image')

  Function Name: 'call_and_return_all_conditional_losses'
    Option #1
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None
    Option #2
      Callable with:
        Argument #1
          image: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='image')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None
    Option #3
      Callable with:
        Argument #1
          image: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='image')
        Argument #2
          DType: bool
          Value: False
        Argument #3
          DType: NoneType
          Value: None
    Option #4
      Callable with:
        Argument #1
          inputs: TensorSpec(shape=(1, 256, 256, 3), dtype=tf.float32, name='inputs')
        Argument #2
          DType: bool
          Value: True
        Argument #3
          DType: NoneType
          Value: None

[hesamsheikh]

Method name is: tensorflow/serving/predict

Thanks. Why doesn't this saved_model not have ordinary attributes of a tensorflow model? Things like summary, train, compile... Is there a workaround for retraining this model?

[hesamsheikh]

I fixed the problem by converting the model to h5. It seems to be okay.

[hesamsheikh]

Is the DetectionOutput layer going to be implemented anytime soon? The current workaround puts considerable calculations on CPU.

Thanks @PINTO0309