How to optimize resnet_v2_50 for tensorflow? (OpenVINO 2018.R3 / R4)

[li__lang]

I have downloaded the model file of resnet_v2_50 from https://github.com/tensorflow/models/tree/master/official/resnet which is saved_model.pb for FP32, but I cannot generate the optimized model by using the following command:

     sudo python mo_tf.py --saved_model_dir /path/to/model_dir  --data_type FP32

And, I get the error information like this:

[Severine_H_Intel]

Dear Lang, 

you must provide an input shape for the model. The MO does not support dynamic input shape ( the -1 ). Add --input_shape <positive_number> to your command.

Best, 

Severine

[li__lang]

Dear Severine，

I test the two commands like：

--input_shape [224,224,3] and --input_shape [1, 224,224,3], 

then I get the following result:

What's wrong?

Best,

Lang

[li__lang]

Can anyone help me？