while converting tensorflow graph(.pb) to openvino IR(.xml), using mo_tf.py, shape of output layer is changing. In my case its changing from [1,13,13,30] to [1,30,13,13]. Can anyone explain this?
Hi shubham. Inference Engine expects NCHW but Tensorflow generates NHWC (channel is last shape element). And this is exactly the switch that you observe.
For more complete information about compiler optimizations, see our Optimization Notice.