Jetson Nano Developer Kit | Getting Started

Easy to run scripts and docs compilation.

Jetson Nano:

• https://developer.nvidia.com/embedded/jetson-nano-developer-kit

Forum with latest news:

• https://forums.developer.nvidia.com/c/agx-autonomous-machines/jetson-embedded-systems/jetson-nano/76

Table of Contents

• ⚙️ Complete official Getting Started
• 🐍 Useful system packages
• 🕯 PyTorch
• ❇️ PyTorch to TensorRT converter
• 🎞 FFmpeg with hardware-accelerated encoding and decoding
• 🔑 SSH Configuration
• 🔮 Disable GUI on boot
• 🗂 Other tutorials

---

⚙️ Complete official Getting Started

Getting Started With Jetson Nano Developer Kit:

• Introduction
• Prepare for Setup
• Write Image to the microSD Card
• Setup and First Boot
• Next Steps
• Troubleshooting

🐍 Useful system packages

sudo apt-get update
sudo apt-get install -y mc git curl htop libcurl4-openssl-dev libssl-dev python3.6-dev python3-pip
sudo apt-get upgrade

# Docker Compose
sudo curl -L "https://github.com/docker/compose/releases/download/1.25.4/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
sudo chmod +x /usr/local/bin/docker-compose

🕯 PyTorch

Topic:

• https://forums.developer.nvidia.com/t/pytorch-for-jetson-nano-version-1-4-0-now-available/72048

pytorch.sh (specify versions at first lines):

#!/usr/bin/env bash
set -e

# Set your python executable
PYTHON="python3.6"
# Set PyTorch version, available: 1.0.0 1.1.0 1.2.0 1.3.0 1.4.0
PYTORCH_VERSION="1.4.0"

PIP="${PYTHON} -m pip"

declare -A TORCHVISION_VERSIONS=(
    ["1.4.0"]="v0.5.0"
    ["1.3.0"]="v0.4.2"
    ["1.2.0"]="v0.4.0"
    ["1.1.0"]="v0.3.0"
    ["1.0.0"]="v0.2.2"
)
TORCHVISION_VERSION=${TORCHVISION_VERSIONS[${PYTORCH_VERSION}]}

declare -A WHEEL_LINKS=(
    ["2.7-1.4.0"]="https://nvidia.box.com/shared/static/1v2cc4ro6zvsbu0p8h6qcuaqco1qcsif.whl"
    ["2.7-1.3.0"]="https://nvidia.box.com/shared/static/6t52xry4x2i634h1cfqvc9oaoqfzrcnq.whl"
    ["2.7-1.2.0"]="https://nvidia.box.com/shared/static/8gcxrmcc6q4oc7xsoybk5wb26rkwugme.whl"
    ["2.7-1.1.0"]="https://nvidia.box.com/shared/static/n9p7u0tem1hqe0kyhjspzz78xpka7f5e.whl"
    ["2.7-1.0.0"]="https://nvidia.box.com/shared/static/d5v4bngglqhdbr4g9ir4eeg6k6miwqnv.whl"
    ["3.6-1.4.0"]="https://nvidia.box.com/shared/static/ncgzus5o23uck9i5oth2n8n06k340l6k.whl"
    ["3.6-1.3.0"]="https://nvidia.box.com/shared/static/phqe92v26cbhqjohwtvxorrwnmrnfx1o.whl"
    ["3.6-1.2.0"]="https://nvidia.box.com/shared/static/06vlvedmqpqstu1dym49fo7aapgfyyu9.whl"
    ["3.6-1.1.0"]="https://nvidia.box.com/shared/static/mmu3xb3sp4o8qg9tji90kkxl1eijjfc6.whl"
    ["3.6-1.0.0"]="https://nvidia.box.com/shared/static/2ls48wc6h0kp1e58fjk21zast96lpt70.whl"
)
declare -A WHEEL_NAMES=(
    ["2.7-1.4.0"]="torch-1.4.0-cp27-cp27mu-linux_aarch64.whl"
    ["2.7-1.3.0"]="torch-1.3.0-cp27-cp27mu-linux_aarch64.whl"
    ["2.7-1.2.0"]="torch-1.2.0a0+8554416-cp27-cp27mu-linux_aarch64.whl"
    ["2.7-1.1.0"]="torch-1.1.0-cp27-cp27mu-linux_aarch64.whl"
    ["2.7-1.0.0"]="torch-1.0.0a0+bb15580-cp27-cp27mu-linux_aarch64.whl"
    ["3.6-1.4.0"]="torch-1.4.0-cp36-cp36m-linux_aarch64.whl"
    ["3.6-1.3.0"]="torch-1.3.0-cp36-cp36m-linux_aarch64.whl"
    ["3.6-1.2.0"]="torch-1.2.0a0+8554416-cp36-cp36m-linux_aarch64.whl"
    ["3.6-1.1.0"]="torch-1.1.0-cp36-cp36m-linux_aarch64.whl"
    ["3.6-1.0.0"]="torch-1.0.0a0+bb15580-cp36-cp36m-linux_aarch64.whl"
)

make_checks() {
    if ! hash ${PYTHON}; then
        echo "Error: ${PYTHON} is not found"
        exit 1
    fi

    PYTHON_VERSION=$(${PYTHON} -V 2>&1 | sed 's/.* \([0-9]\).\([0-9]\).*/\1.\2/')
    if [[ "$PYTHON_VERSION" != "2.7" ]] && [[ "$PYTHON_VERSION" != "3.6" ]]; then
        echo "Error: this script requires Python 2.7 or 3.6"
        exit 2
    fi
}

install_dependencies() {
    sudo apt-get install -y libopenblas-base libjpeg-dev zlib1g-dev
    eval "${PIP} install Cython numpy pillow\<7"
}

install_pytorch() {
    cd /tmp
    key="${PYTHON_VERSION}-${PYTORCH_VERSION}"
    wget ${WHEEL_LINKS[$key]} -O ${WHEEL_NAMES[$key]}
    eval "${PIP} install ${WHEEL_NAMES[$key]}"
}

install_torchvision() {
    cd /tmp
    git clone --branch ${TORCHVISION_VERSION} https://github.com/pytorch/vision torchvision
    cd torchvision
    sudo ${PYTHON} setup.py install
}

# Run
make_checks
install_dependencies
install_pytorch
install_torchvision

❇️ PyTorch to TensorRT converter

Tool for converting PyTorch models to TensorRT:

• https://github.com/NVIDIA-AI-IOT/torch2trt

torch2trt.sh:

#!/usr/bin/env bash
set -e

# Set your python executable
PYTHON="python3.6"

cd /tmp

git clone https://github.com/NVIDIA-AI-IOT/torch2trt
cd torch2trt
sudo ${PYTHON} setup.py install

🎞 FFmpeg with hardware-accelerated encoding and decoding

Accelerated encoding of H.264 and HEVC and decoding of VP8, VP9, MPEG2, H.264 and HEVC.

Repo: jocover / jetson-ffmpeg

ffmpeg.sh:

#!/usr/bin/env bash
set -e

cd /tmp

# Build and install library for ffmpeg
git clone https://github.com/jocover/jetson-ffmpeg.git
cd jetson-ffmpeg
mkdir build
cd build
cmake ..
make -j6
sudo make install
sudo ldconfig

# Build and install ffmpeg
git clone git://source.ffmpeg.org/ffmpeg.git -b release/4.2 --depth=1
cd ffmpeg
wget https://github.com/jocover/jetson-ffmpeg/raw/master/ffmpeg_nvmpi.patch
git apply ffmpeg_nvmpi.patch
./configure --enable-nvmpi
make -j6
sudo make install

Usage examples:

• ffmpeg -c:v h264_nvmpi -i input_file -f null -
• ffmpeg -i input_file -c:v h264_nvmpi <output.mp4>

🔑 SSH Configuration

/etc/ssh/sshd_config:

PasswordAuthentication no
PubkeyAuthentication yes

🔮 Disable GUI on boot

Good choice for ssh usage. It saves an average 1 GB of RAM (of 4).

sudo systemctl set-default multi-user.target

and reboot. Source.

---

🗂 Other tutorials

• https://elinux.org/Jetson_Zoo
  • TensorFlow with Keras
  • MXNet
  • ROS
  • NVIDIA Isaac SDK
  • ... and others