jetson NANO tensorflow pb to tensorRT problem

I am using the jetson NANO with its official image (ubuntu 18.04), and tensorflow (1.13.0).

I tried to retrain an efficient net model in keras and converted the model into tensorflow pb file using this tool:
https://github.com/amir-abdi/keras_to_tensorflow

The script and original tensorflow pb file is here: GitHub - captainst/tensorflow-to-tensorRT-convert: A conversion test from tensorflow freezed pb to tensorRT format

Some of the warnings / errors I got are:

I tensorflow/contrib/tensorrt/convert/convert_graph.cc:928] Number of TensorRT candidate segments: 150
I tensorflow/contrib/tensorrt/convert/convert_graph.cc:1030] TensorRT node TRTEngineOp_0 added for segment 0 consisting of 5 nodes succeeded.
W tensorflow/contrib/tensorrt/convert/convert_nodes.cc:3743] Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,192,?,?] has an unknown non-batch dimension at dim 2
W tensorflow/contrib/tensorrt/convert/convert_graph.cc:1036] TensorRT node TRTEngineOp_1 added for segment 1 consisting of 5 nodes failed: Invalid argument: Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,192,?,?] has an unknown non-batch dimension at dim 2. Fallback to TF...
W tensorflow/contrib/tensorrt/convert/convert_nodes.cc:3743] Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,32,?,?] has an unknown non-batch dimension at dim 2
W tensorflow/contrib/tensorrt/convert/convert_graph.cc:1036] TensorRT node TRTEngineOp_2 added for segment 2 consisting of 5 nodes failed: Invalid argument: Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,32,?,?] has an unknown non-batch dimension at dim 2. Fallback to TF...
W tensorflow/contrib/tensorrt/convert/convert_nodes.cc:3743] Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,192,?,?] has an unknown non-batch dimension at dim 2
W tensorflow/contrib/tensorrt/convert/convert_graph.cc:1036] TensorRT node TRTEngineOp_3 added for segment 3 consisting of 5 nodes failed: Invalid argument: Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,192,?,?] has an unknown non-batch dimension at dim 2. Fallback to TF...
W tensorflow/contrib/tensorrt/convert/convert_nodes.cc:3743] Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,192,?,?] has an unknown non-batch dimension at dim 2
W tensorflow/contrib/tensorrt/convert/convert_graph.cc:1036] TensorRT node TRTEngineOp_4 added for segment 4 consisting of 5 nodes failed: Invalid argument: Validation failed for TensorRTInputPH_0 and input slot 0: Input tensor with shape [?,192,?,?] has an unknown non-batch dimension at dim 2. Fallback to TF...

Almost all segments, except for 2, are failed.The generated tensorRT model works. However it does not show a noticeable difference from the original tensorflow pb model.

To give it a quick fix, try to add argument

is_dynamic_op=True

to create_inference_graph.
Reference: https://medium.com/tensorflow/high-performance-inference-with-tensorrt-integration-c4d78795fbfe

However, the generated model does not give a noticeable improvement either.