Not able to Optimize re-trained SSD-MobileNet-v2

Ramesh__Siddharth · ‎12-08-2019

I have trained re-trained the SSD-MobileNet-v2 model on my custom dataset with tensorflow-GPU=1.11* and object detection API v1.10*
Im unable to convert the obtained frozen graph to Intermediate Representation (.xml,.bin). The following is the error trace when i run mo_tf.py

'''''''''''''''''''''''''''''''''
mo_tf.py --input_model frozen_inference_graph.pb --tensorflow_use_custom_operations_config ssd_support_api_v1.14.json --tensorflow_object_detection_api_pipeline_config pipeline.config

Model Optimizer arguments:
Common parameters:
   - Path to the Input Model:    /frozen_inference_graph.pb
   - Path for generated IR:    /opt/intel/openvino_2019.3.376/deployment_tools/model_optimizer/.
   - 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 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:    /pipeline.config
   - Operations to offload:    None
   - Patterns to offload:    None
   - Use the config file:    /opt/intel/openvino_2019.3.376/deployment_tools/model_optimizer/extensions/front/tf/ssd_support_api_v1.14.json
Model Optimizer version:    2019.3.0-408-gac8584cb7
The Preprocessor block has been removed. Only nodes performing mean value subtraction and scaling (if applicable) are kept.
[ ERROR ] Failed to match nodes from custom replacement description with id 'ObjectDetectionAPISSDPostprocessorReplacement':
It means model and custom replacement description are incompatible.
Try to correct custom replacement description according to documentation with respect to model node names
[ ERROR ] Cannot infer shapes or values for node "Postprocessor/Cast_1".
[ ERROR ] 0
[ ERROR ]
[ ERROR ] It can happen due to bug in custom shape infer function <function Cast.infer at 0x7f7b87bc0048>.
[ 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 ] 0
Stopped shape/value propagation at "Postprocessor/Cast_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.
Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.middle.PartialInfer.PartialInfer'>): 0
Stopped shape/value propagation at "Postprocessor/Cast_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.
'''''''''''''''''''''''''''''''''''

PS:
i) I have appended the start_points in "ssd_support_api_v1.14.json" with "Postprocessor/Cast_1" and yet I receive this error.
ii) I was able to successfully convert the "frozen_graph.pb" file from the SSD-MobileNet-v2 saved model available from the official tensorflow repo. (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md)

Is there any particular commit of the object_detection repository on which we should train? I have successfully attempted training on the latest object detection API (2019) with tensorflow 1.15 and 1.11
Both the trained frozen graph prompted the above error message.

Please do suggest a workaround for this error or point me to the particular version/commit of the object_detection API where I could successfully optimize a retrained SSD-MobileNet-v2 model.

Luis_at_Intel · ‎12-10-2019

Hi Siddharth,

Thanks for reaching out. I see you are trying to convert SSD-MobileNet-V2 model to IR, could you try using the ssd_v2_support.json instead of ssd_v2_support.json and see if that works?

I can also attempt to convert your re-trained model and see if I can find a solution, if possible share your model files so we can try. You can share privately these files in case you don't want to share them publicly.

Regards,

Luis