How to run SSD Mobilenet V2 object detection on Jetson Nano at 20+ FPS | DLology

First, make sure you have flashed the latest JetPack 4.3 on your Jetson Nano development SD card.

Run the docker

docker run --runtime nvidia --network host --privileged -it docker.io/zcw607/trt_ssd_r32.3.1:0.1.0

Then run this command to benchmark the inference speed.

python3 trt_ssd_benchmark.py

To build the docker yourself

To build the docker on Jetson nano or any other arm64 board.

docker build -t docker.io/<your user name>/trt_ssd_r32.3.1:0.1.0 . -f ./Dockerfile
docker push docker.io/<your user name>/trt_ssd_r32.3.1:0.1.0

To build the docker on a x86 machine, requires podman and qemu installed.

podman build -v /usr/bin/qemu-aarch64-static:/usr/bin/qemu-aarch64-static -t docker.io/<your user name>/trt_ssd_r32.3.1:0.1.0 . -f ./Dockerfile
podman push docker.io/<your user name>/trt_ssd_r32.3.1:0.1.0