How to convert ShuffleNet by using Model Optimizer?

[He__Jinyang1]

  Hello, everyone! I want to convert my ShuffleNet by using Model Optimizer. There is something wrong:

 [ ERROR ]  Exception occurred during running replacer "None" (<class 'extensions.front.tf.assign_elimination.AssignSubElimination'>): Data flow edge coming out of AssignSub node tower_0/ShuffleNetV2/Stage3/unit_7/depthwise/BatchNorm/AssignMovingAvg_1

I guess OpenVINO don't support some operation or node?  what should I do for converting the  ShuffleNet model? Thanks!!!!

[Sahira_Intel]

Hi Jinyang,

What version of Shufflenet are you using? OpenVINO only supports Shufflenet V2 on Caffe - please make sure you are using the correct version. To convert your model to IR, see here. 

Please let me know if this information is helpful. 

Best Regards,

Sahira 

[He__Jinyang1]

Sahira R. (Intel) wrote:

Hi Jinyang,

What version of Shufflenet are you using? OpenVINO only supports Shufflenet V2 on Caffe - please make sure you are using the correct version. To convert your model to IR, see here. 

Please let me know if this information is helpful. 

Best Regards,

Sahira 

Thanks for your help Sahira! Actually, I want to convert a trained Shufflenet V2 network which is used to detect facial landmark. My network is trained by TensorFlow. I want to add the facial landmark network to the OpenVINO-Driver-Behaviour project for facial analysis.

[Sahira_Intel]

Hi Jinyang,

Shufflenet  V2 is currently not a supported topology by Tensorflow (only Caffe and ONNX) - see here for a list of supported Tensorflow topologies. 

Please let me know if you have any further questions.

Best Regards,

Sahira