Error in generating IR files for Faster R-CNN RPN layers

[Vihari_G_Intel]

Hi,

I am trying to generate IR files for Faster R-CNN model with DenseNet feature extractor. I tried it on openvino versions 2019 R1,R1.1 and 2018 R4, R5 . 

Command: 

python mo_tf.py

--input_model ~\Desktop\tf_1.121\frozen_inference_graph.pb

--tensorflow_use_custom_operations_config extensions\front\tf\faster_rcnn_support_api_v1.10.json 

--tensorflow_object_detection_api_pipeline_config ~\Desktop\tf_1.121\pipeline.config

--output_dir ~\Desktop\tf_1.121\output

Error:

Model Optimizer arguments:

Common parameters:

    - Path to the Input Model:      ~\Desktop\tf_1.121\frozen_inference_graph.pb

    - Path for generated IR:        ~\Desktop\tf_1.121\output

    - IR output name:       frozen_inference_graph

    - Log level:    ERROR

    - Batch:        Not specified, inherited from the model

    - Input layers:         Not specified, inherited from the model

    - Output layers:        Not specified, inherited from the model    

    - Input shapes:         Not specified, inherited from the model

    - Mean values:  Not specified

    - Scale values:         Not specified- Scale values:         Not specified

    - Scale factor:         Not specified

    - Precision of IR:      FP32

    - Enable fusing:        True

    - Enable grouped convolutions fusing:   True

    - Move mean values to preprocess section:       False

    - Reverse input channels:       False

TensorFlow specific parameters:

    - Input model in text protobuf format:  False

    - Path to model dump for TensorBoard:   None

    - List of shared libraries with TensorFlow custom layers implementation:        None

    - Update the configuration file with input/output node names:   None

    - Use configuration file used to generate the model with Object Detection API:  ~\Desktop\tf_1.121\pipeline.config

    - Operations to offload:        None

    - Patterns to offload:  None

    - Use the config file:  C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\extensions\front\tf\faster_rcnn_support_api_v1.10.json

Model Optimizer version:        2019.1.0-341-gc9b66a2

The Preprocessor block has been removed. Only nodes performing mean value subtraction and scaling (if applicable) are kept.

[ ERROR ]  Exception occurred during running replacer "ObjectDetectionAPIDetectionOutputReplacement" (<class 'extensions.front.tf.ObjectDetectionAPI.ObjectDetectionAPIDetectionOutputReplacement'>): Found the following nodes '[]' with name 'crop_proposals' but there should be exactly 1. Looks like ObjectDetectionAPIProposalReplacement replacement didn't work.

Tried with both faster_rcnn_support_api_v1.7.json and faster_rcnn_support.json.  and  I can see all the nodes given in these JSONs in the graph.

Waiting to hear back.

Thanks.

[Shubha_R_Intel]

Dear Gandrakota, Vihari,

Please see my answer to this post

I'm assuming your model is custom trained ?

Thanks

Shubha