ssd-mobilenet-v1 can't convert to TensorRT in current repo

System information

• What is the top-level directory of the model you are using: object_detection
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 18.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): 1.13
• Bazel version (if compiling from source): No
• CUDA/cuDNN version: 10
• GPU model and memory: GTX-1050Ti 4G、Jetson Nano
• Exact command to reproduce:

Describe the problem

I train the ssd-mobilenet-v1 model by using object_detection api (current version is #d735ecf83a99545e46ff27744b4423379ddbb262) and export it, when I try to use tensorrt code (provided by nvidia) to convert to TensorRT, it told me "Unsupported operation: _Cast"

I reset the object_detection api version to #ae0a9409212d0072938fa60c9f85740bb89ced7e, use the same checkpoints and export the frozen_model again and use the same code try to convert it to TensorRT, it succeed.

It seems that current add the extra layer called _Cast and TensorRT can't convert it.

Source code / logs

https://devtalk.nvidia.com/default/topic/1055295/jetson-nano/fine-tune-ssd_mobilenet_v1-model-can-t-convert/post/5355581/#5355581

https://devtalk.nvidia.com/default/topic/1043557/tensorrt/error-uffparser-parser-error-boxpredictor_0-reshape-reshape-1-dimension-specified-more-than-1-/post/5317840/#5317840