EfficientDet python sample script: create_onnx.py fail for EfficientDetD4 checkpoint

Description

/TensorRT/samples/python/efficientdet/python/create_onnx.py sample script fails in update_shapes() function for checkpoint created via finetuning an efficientdet-d4 pretrained model (no error when repeating process based on efficientdet-d7 model).

Traceback (most recent call last):
  File "create_onnx.py", line 453, in <module>
    main(args)
  File "create_onnx.py", line 424, in main
    effdet_gs.update_shapes()
  File "create_onnx.py", line 228, in update_shapes
    scale_h = concat.inputs[1].values[0] / node.inputs[0].shape[2]
TypeError: ufunc 'divide' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule ''safe''

I found that the node.inputs[0].shape[2] is of size None

the following (hacky) code change enabled me to get past this problem and i successfully created an engine from the onnx file created. However, i do not know if my code change resulted in the 'optimal' graph.

Environment

**Container **: nvcr.io/nvidia/tensorflow:22.06-tf2-py3
all of the following libraries exist by default the container specified

TensorRT Version: libnvinfer-bin 8.2.5-1+cuda11.4
NVIDIA GPU: 3080ti
NVIDIA Driver Version: 510.73.05
CUDA Version: Build cuda_11.7.r11.7/compiler.31442593_0
CUDNN Version: libcudnn8 8.4.1.50-1+cuda11.6
Operating System: ubuntu 18.04
Python Version (if applicable): 3.8.10
Tensorflow Version (if applicable): 2.8
PyTorch Version (if applicable): N/A

Steps To Reproduce

option A:

1. git clone https://github.com/ece85/Deploy_EfficientDet.git
2. run setup_and_repro.sh (this will build a docker image and run a script to reproduce the problem).

option B:

if you are confident that you current dev environment is similar enough to mine then just download the folder linked by :

https://drive.google.com/drive/folders/1yWTzCxeDUiM93Se4tOL29c7VBnuCUuHN?usp=sharing

and point that to the following python script call:

cd /TensorRT/samples/python/efficientdet

python3 create_onnx.py \
    --input_size 1024,1024 \
    --saved_model $saved_model \ 
    --onnx model.onnx

where $saved_model points to folder downloaded from link.