DEEPX SDK Tutorials

Welcome! This repository is a collection of hands-on JupyterLab tutorials for the DEEPX SDK (DX-All Suite). It starts with installing the SDK, continues with the compiler and the runtime, and ends with complete Python and C++ applications on the DEEPX NPU.

These tutorials target the dx-all-suite release tag v2.4.2 (the default in Tutorial 01 section 1.1.2): DX-COM 2.4.1, DX-RT 3.4.2, DX-APP 3.2.2, DX-STREAM 3.1.2. The same component versions were validated on main as of September 2026 (v2.4.3).

Follow the tutorials in order: 00 and 01 first, then 02 to 06, then the demos. Each notebook ends with a Next pointer to the one that follows.

dx-tutorials
├── notebooks
│   ├── T00-JupyterLab-QuickStart/        jupyterlab_quickstart.ipynb
│   ├── T01-Getting-Started/              getting_started.ipynb
│   ├── T02-DX-APP/                       dx_app.ipynb
│   ├── T03-E2E-AI-Workflow/              e2e_ai_workflow.ipynb
│   ├── T04-DX-STREAM/                    dx_stream.ipynb
│   ├── T05-DX-Compiler/                  dx_com_01_beginner / 02_intermediate / 03_advanced.ipynb
│   ├── T06-DX-Runtime/                   dx_rt_01_beginner / 02_intermediate / 03_advanced.ipynb
│   ├── T10-demo-paddleocr/               paddleocr_v6.ipynb, app/ (Python application)
│   ├── T20-demo-yolo-multi/              yolo_multi.ipynb, app/ (C++), get_resources.sh
│   ├── T21-demo-yolo26-od-pose-seg-depth/ yolo26_od_pose_seg_depth.ipynb, app/, get_resources.sh
│   ├── T22-demo-clip-single/             clip_single.ipynb, app/, get_resources.sh
│   ├── T23-demo-hand-landmarks/          hand_landmarks.ipynb, app/, get_resources.sh
│   └── T24-demo-pidnet-cityscapes/       pidnet_cityscapes.ipynb, app/, get_resources.sh
├── docs/                   platform notes (Raspberry Pi 5, Orange Pi 5 Plus, Intel GPU) and the maintenance checklist
├── tutorial_paths.py       shared path resolution and requirement checks used by every notebook
├── config.example.json     template for the user-local config.json (see below)
├── requirements.txt        JupyterLab environment
├── run-jupyter-lab.sh      creates the environment and starts JupyterLab
└── sudo_no_password.sh     optional: lets the install cells run sudo without a password (see below)

Every tutorial keeps its generated files (downloaded models, compiled DXNN files, build outputs, reports) under its own workspace/ or assets/ directory. Those directories are ignored by git.

sudo apt-get update
sudo apt-get install python3-venv build-essential python3-dev git-all ffmpeg tree curl cmake

curl -LsSf https://astral.sh/uv/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"
uv --version

git clone --depth 1 --single-branch https://github.com/DEEPX-AI/dx-tutorials.git
cd dx-tutorials

./run-jupyter-lab.sh

The script creates .venv with the packages from requirements.txt and starts JupyterLab from it without activating the environment. Notebook cells and the JupyterLab terminal therefore see the system python3, so the SDK installers create their own Python environments instead of touching this one. Packages that a single tutorial needs (ONNX tools, OpenCV, PyTorch for the export examples) are installed by that notebook with uv pip install --python "{sys.executable}" when they are first used.

Tutorial 01 clones DX-All Suite and installs the SDK. The location is resolved in this order and shown at the top of every notebook:

1. the DX_ALL_SUITE_DIR environment variable,
2. config.json in this directory (written by Tutorial 01; config.example.json shows the fields; the file itself is not tracked by git),
3. auto-detection of ~/dx-all-suite, a dx-all-suite folder next to dx-tutorials, or ~/Works/dx-all-suite,
4. the default ~/dx-all-suite.

To set it from a terminal:

python tutorial_paths.py --set ~/my/dx-all-suite
python tutorial_paths.py --show

The SDK installers call sudo apt-get. A notebook cell cannot answer a password prompt, so the install cells check sudo -n true first: when sudo works without a password they run the installer inline, otherwise they print the same commands for a terminal. sudo_no_password.sh configures passwordless sudo for the sudo group by writing /etc/sudoers.d/sudo-nopasswd. Use it only on a personal or lab machine; it removes the password check for every sudo command of that group.

• The first code cell of every notebook is identical: it loads tutorial_paths.py, resolves the SDK location, checks the requirements of that notebook, and prints a status table.
• Every command a learner needs appears verbatim in the cell that runs it (!git clone ..., !dxcom ..., !./build.sh), with the same command shown in the Markdown above it so it can be copied into a terminal. The only exceptions are a few cells that parse a command's output; those print the exact command with a $ prefix before running it.
• Cells that open a window or need hardware are guarded by a flag (RUN_DXTRON, RUN_DEMOS, RUN_CAMERA, SHOW_WINDOW, HEADLESS, FLASH_FIRMWARE) so that Run All never blocks or changes the system by accident.
• Terminal commands in the Markdown write <DX_ALL_SUITE_DIR> for the DX-All Suite directory (default ~/dx-all-suite; Tutorial 01 section 1.1 prints the configured value) and <workspace> for the tutorial's own workspace/ directory. The code cells substitute the real paths.
• Sections are numbered ## 1., ### 1.1, #### 1.1.1. Each notebook starts with learning objectives and prerequisites and ends with troubleshooting, a summary, a completion checklist, and a pointer to the next tutorial.

Every tutorial notebook (T00 to T06, T10, and T20 to T24) is available in English, Korean, Japanese, and Chinese. The English notebook is the source of truth; the translated copies are generated next to it and share the same code cells, so one verification run covers every language. Each notebook starts with a language switcher line (🌐 English | 한국어 | 日本語 | 中文); click a language to open that copy.

# After editing English Markdown or a translation: regenerate every <lang>.ipynb
python scripts/build_i18n.py build

# Report stale or missing translations and out-of-date generated notebooks (exit 1 on problems)
python scripts/build_i18n.py check

# Find blocks that still contain the English text
python scripts/i18n_untranslated.py

# Start a new language for one notebook: write an English template to translate
python scripts/build_i18n.py init notebooks/T02-DX-APP/dx_app.ipynb --lang de

A translation block records the hash of the English cell it was made from. When the English text changes, check lists that cell as stale; translate it again and run python scripts/build_i18n.py stamp <notebook> to record the new hash.

Platform notes: Raspberry Pi 5 PCIe Gen3, Orange Pi 5 Plus kernel headers, Intel GPU drivers for DX-STREAM.

For questions or feedback, please contact dgkim@deepx.ai or create an issue ticket.