This demo application shows how to create a simple Keras model and optimize it using the TF-TRT runtime on the Jetson platform.
Flash a Jetson Nano with JetPack 4.4 by following the instructions here. Log in and run the following commands:
git clone https://github.com/pascal-hwky/tftrt-jetson.git
cd tftrt-jetson
./arm.sh
A built-in toy model is used. If you would like to use your own SavedModel, download it to a folder model
in this repo:
aws s3 sync s3://path/to/saved/model model
Then, run arm.sh
again.
If you want to make changes to the code while testing in a Docker container on a development PC, run the following command:
./dev.sh
This command will map the current directory to /shared/
in the container. To run the code, execute the following command in the container:
python3 main.py