Tensorflow model exporter error - Cannot infer shapes or values for node "image_padding/Ceil_1"

[Moosa__Azmath]

I am trying to export a frozen inference graph pb file.  It is a custom bbox detection network with inception v1 blocks.  While running the model optimizer, I am getting some errors.  The graph was trained and exported from tf v1.10.  I want to know how I  can debug this model and get it exported.

The command line I used

/opt/intel/openvino/deployment_tools/model_optimizer/mo_tf.py --input_model /data/work/face-rec/face-rec/src/assets/faceboxes-model.pb --input_shape "(1,1024,1024,3)" --input image_tensor  --output_dir ./ --log_level=DEBUG

The error that I am getting is as follows:

[ 2020-04-22 10:52:33,606 ] [ DEBUG ] [ infer:128 ]  --------------------
[ 2020-04-22 10:52:33,606 ] [ DEBUG ] [ infer:129 ]  Partial infer for image_padding/Ceil_1
[ 2020-04-22 10:52:33,606 ] [ DEBUG ] [ infer:130 ]  Op: Ceil
[ 2020-04-22 10:52:33,606 ] [ DEBUG ] [ infer:131 ]  Inputs:
[ 2020-04-22 10:52:33,606 ] [ DEBUG ] [ infer:31 ]  input[0]: shape = [], value = 8.0
[ INFO ]  Called "tf_native_tf_node_infer" for node "image_padding/Ceil_1"
[ 2020-04-22 10:52:33,607 ] [ DEBUG ] [ tf:226 ]  Added placeholder with name 'image_padding/truediv_1_port_0_ie_placeholder'
[ 2020-04-22 10:52:33,607 ] [ DEBUG ] [ tf:239 ]  update_input_in_pbs: replace input 'image_padding/truediv_1' with input 'image_padding/truediv_1_port_0_ie_placeholder'
[ 2020-04-22 10:52:33,607 ] [ DEBUG ] [ tf:247 ]  Replacing input '0' of the node 'image_padding/Ceil_1' with placeholder 'image_padding/truediv_1_port_0_ie_placeholder'
[ ERROR ]  Cannot infer shapes or values for node "image_padding/Ceil_1".
[ ERROR ]  Input 0 of node image_padding/Ceil_1 was passed float from image_padding/truediv_1_port_0_ie_placeholder:0 incompatible with expected double.
[ ERROR ]  
[ ERROR ]  It can happen due to bug in custom shape infer function <function tf_native_tf_node_infer at 0x7f4cceef7378>.
[ 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).
[ 2020-04-22 10:52:33,759 ] [ DEBUG ] [ infer:196 ]  Node "image_padding/Ceil_1" attributes: {'pb': name: "image_padding/Ceil_1"
op: "Ceil"
input: "image_padding/truediv_1_port_0_ie_placeholder"
attr {
  key: "T"
  value {
    type: DT_DOUBLE
  }
}
, '_in_ports': {0: {'control_flow': False}}, '_out_ports': {0: {'control_flow': False}}, 'kind': 'op', 'name': 'image_padding/Ceil_1', 'op': 'Ceil', 'infer': <function tf_native_tf_node_infer at 0x7f4cceef7378>, 'is_output_reachable': True, 'is_undead': False, 'is_const_producer': False, 'is_partial_inferred': False}
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.middle.PartialInfer.PartialInfer'>): Stopped shape/value propagation at "image_padding/Ceil_1" node. 
 For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #38. 
[ 2020-04-22 10:52:33,761 ] [ DEBUG ] [ main:317 ]  Traceback (most recent call last):
  File "/home/azmath/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/importer.py", line 426, in import_graph_def
    graph._c_graph, serialized, options)  # pylint: disable=protected-access
tensorflow.python.framework.errors_impl.InvalidArgumentError: Input 0 of node image_padding/Ceil_1 was passed float from image_padding/truediv_1_port_0_ie_placeholder:0 incompatible with expected double.

I think I need to override the DT_DOUBLE to Float.  Could anyone guide me on how I can do that?

[SuryaPSC_Intel]

Hi Azmath,

Can you please confirm if the model is one of the supported topologies of tensorflow. For face detection problem kindly refer to Interactive face detection demo.

Best Regards,

Surya

[ashna_12]

Is this issue solved? If yes, how did you solve this?

I'm also facing a similar kind of issue.