Model Optimizer - converting ONNX models

[Andres_M_Intel2]

I am trying to convert ONNX models using Model Optimizer.

I only succeeded to convert 3 of the 8 possible models (bvlc_googlenet, inception_v1, squeezenet) that should be covered (openVINO 2018 R2).

1) I wonder if you have a BKM similar to this one ? https://software.intel.com/en-us/articles/OpenVINO-Using-MXNet
2) Could you describe the 8 topologies that are supported ?

Please find the output from the basic command i am using:

>> python3 /opt/intel/computer_vision_sdk/deployment_tools/model_optimizer/mo.py --input_model model.onnx --output_dir ./ --framework onnx
Model Optimizer arguments:
Common parameters:
 - Path to the Input Model:  onnx/tiny_yolov2/model.onnx
 - Path for generated IR:  onnx/tiny_yolov2/./
 - IR output name:  model
 - 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
ONNX specific parameters:
Model Optimizer version:  1.2.110.59f62983
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  FusedBatchNorm doesn't support is_test=False
[ ERROR ]  Cannot infer shapes or values for node "scalerPreprocessor".
[ ERROR ]  There is no registered "infer" function for node "scalerPreprocessor" with op = "ImageScaler". Please implement this function in the extensions.
 For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #37.
[ ERROR ]
[ ERROR ]  It can happen due to bug in custom shape infer function <UNKNOWN>.
[ 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 ]  Stopped shape/value propagation at "scalerPreprocessor" node.
 For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #38.

Thanks in advance,
Andres

[Severine_H_Intel]

Dear Andres, 

ONNX support is still in feature preview in R2, which explains the few documentation about it. 

Have you downloaded the models from our documentation? computer_vision_sdk_2018.2.299/deployment_tools/documentation/Intro.html

If not and if you have downloaded the models from the ONNX github, you should pay attention to use models with opset under 7. If I take the example of ResNet50, you will see a list like the one below and you should pick one of the first two models. We do not support yet opset 7 and above. This is true for any ONNX models on the github page. 

• release 1.1: https://s3.amazonaws.com/download.onnx/models/opset_3/resnet50.tar.gz
• release 1.1.2: https://s3.amazonaws.com/download.onnx/models/opset_6/resnet50.tar.gz
• release 1.2: https://s3.amazonaws.com/download.onnx/models/opset_7/resnet50.tar.gz
• master: https://s3.amazonaws.com/download.onnx/models/opset_8/resnet50.tar.gz

Best, 

Severine