🚀 Jetson Orin Nano — Dev Environment Setup

🌐 繁體中文

An interactive Bash script for setting up a full deep learning stack on Jetson Orin Nano.

• Target: JetPack 6.2 / L4T R36.4.x / R36.5.x
• Python: 3.10 (Miniconda, conda env)

---

📋 Requirements

Item	Requirement
JetPack	6.2 (L4T R36.4.x / R36.5.x)
Swap	≥ 8 GB (required for OpenCV CUDA build)
Free disk	≥ 20 GB

---

📦 Wheel Files

Download the following .whl files from Jetson AI Lab and place them in ~/packages/jetson_wheels/.

Or download directly from Google Drive: 📁 Jetson Wheels

• PyTorch
• Torchvision
• ONNX Runtime GPU
• CUDA Python
• CuPy (CUDA 12x)

cv2 and tensorrt do not need to be downloaded. The script automatically creates symlinks from the system path.

---

🚀 Quick Start

chmod +x jetson_setup.sh

# Optional: set a custom conda env name (default: tools)
export CONDA_ENV=tools

bash jetson_setup.sh

---

🛠 Menu

p) 🔍 Preflight check  (wheels / swap / disk)
1) 🔧 System base      (snap fix + Chinese input)
2) 🔄 System update    (apt update + jtop fix)
3) 📷 OpenCV CUDA      (full rebuild, ~2hr)
4) 🐍 Conda env        (create env + symlinks)
5) 📦 Package install  (wheels + pip)
v) ✅ Validate         (check packages + GPU)
a) ⚡ Run all          (p → 1 → 2 → 3 → 4 → 5 → v)
q) 👋 Quit

Recommended for a fresh device: run p first to verify the environment, then a for full install.

---

⚙️ Configuration

Variable	Default	Description
CONDA_ENV	tools	Conda environment name
PACKAGES_DIR	~/packages/jetson_wheels	Wheel directory
OPENCV_SCRIPT_SHA256	(empty)	SHA256 of the OpenCV build script

---

🐛 Troubleshooting

cv2 has no CUDA support

python -c "import cv2; print(cv2.__file__)"
# Should point to /usr/lib/python3.10/dist-packages/cv2/...
# If not, re-run step 4

tensorrt import fails

export LD_LIBRARY_PATH=/usr/lib/aarch64-linux-gnu:$LD_LIBRARY_PATH