DEEPX — Raspberry Pi 5 Examples

Compact, practical AI inference examples for the Raspberry Pi 5 using the DEEPX NPU. Examples range from object detection and pose estimation to instance segmentation and image denoising — designed to get you running on edge AI quickly.

Quick links

• Sixfab Website | Sixfab Docs
• DEEPX Website
• Runtime & tools: sixfab-dx dx_rt
• Driver: dx_rt_npu_linux_driver
• Full toolchain: dx-all-suite

---

Repository Layout

deepx-rpi5-examples/
├── auto-install.sh           # One-shot setup: resources, Python deps, C++ build
├── python_examples/          # Full Python demo collection (28 demos, TUI launcher)
├── cpp_examples/             # Full C++ demo collection (26 demos, TUI launcher)
├── community_projects/       # Community contributions and templates
└── docs/                     # Guides for installation and usage

---

Getting Started

Prerequisites

Install the DEEPX runtime first via the Sixfab APT repository:

Install the apt-repo-sixfab package

wget https://github.com/sixfab/sixfab_dx/releases/download/v0.1/apt-repo-sixfab.deb
sudo dpkg -i apt-repo-sixfab.deb

Install sixfab-dx package

The sixfab-dx package automatically installs the DEEPX runtime on the system.

sudo apt update && sudo apt install sixfab-dx

Enable PCIe Gen 3 for optimal performance

To get the best inference throughput from the DX-M1 NPU, open the boot configuration file:

sudo nano /boot/firmware/config.txt

Add the following lines at the end of the file:

dtparam=pciex1
dtparam=pciex1_gen=3

Save and exit (Ctrl+X, then Y, then Enter), then reboot:

sudo reboot

Verify the NPU is recognized

dxrt-cli -s

See docs/install-raspberry-pi5.md for full hardware and software setup instructions.

Quick Setup

Once sixfab-dx is installed, run the setup script to download models and build the C++ examples:

git clone https://github.com/sixfab/sixfab-dx-examples.git
cd sixfab-dx-examples
./auto-install.sh

auto-install.sh will:

1. Validate the DEEPX runtime is working
2. Download models, videos, and sample images from the GitHub release
3. Install Python dependencies into the Sixfab venv
4. Build all C++ demos

After setup, activate the environment for every new session:

source /opt/sixfab-dx/venv/bin/activate

After setup, activate the environment for every new session:

source /opt/sixfab-dx/venv/bin/activate

Note: This activates the shared Sixfab virtual environment. If you prefer to use your own isolated environment, you can create one and install the required wheels manually:

python3 -m venv ~/my-venv
source ~/my-venv/bin/activate

Pre-built wheels for dx_engine and numpy are available under /opt/sixfab-dx/wheels/:

ls /opt/sixfab-dx/wheels/
# dx_engine-3.3.0-cp311-cp311-linux_aarch64.whl
# dx_engine-3.3.0-cp313-cp313-linux_aarch64.whl
# numpy-2.4.4-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
# numpy-2.4.4-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl

Install the wheels matching your Python version (cp311 → Python 3.11, cp313 → Python 3.13):

pip install /opt/sixfab-dx/wheels/numpy-2.4.4-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl
pip install /opt/sixfab-dx/wheels/dx_engine-3.3.0-cp311-cp311-linux_aarch64.whl

To check your Python version: python3 --version

---

Example Collections

Python Examples (python_examples/)

A full collection of 28 demos with a Rich-based TUI launcher. Covers object detection (13 YOLO variants, SCRFD, YOLOX), segmentation, classification, pose estimation, PPU-accelerated variants, and advanced analytics (zone intrusion, people tracking, traffic counting, queue analysis).

# Launch the interactive menu
bash python_examples/start.sh

See docs/python-examples.md for the full demo list and configuration guide.

C++ Examples (cpp_examples/)

26 demos written in idiomatic C++17 using only OpenCV and the DeepX SDK — no Qt, no YAML. Same categories as the Python collection. Each demo is a single readable file.

# Launch the interactive menu
bash cpp_examples/start.sh

# Or run a binary directly
./cpp_examples/build/object_detection/yolov8_demo --source libcamera

See docs/cpp-examples.md for the demo list, build instructions, and flags.

---

Input Sources

All demos accept the same set of input sources:

Source	config.yml value
Raspberry Pi camera	rpicam
USB webcam	webcam
Video file	video
Single image	image
RTSP stream	rtsp

---

Demo Gallery

YOLOv8 — object detection	YOLO Pose — pose estimation
YOLOv8 Instance Segmentation	DeepLabV3 Semantic Segmentation
SCRFD Face Detection	YOLOv5 — PPU accelerated
YOLOv8 — PPU accelerated	YOLOv8 — Async inference
OSNet Re-ID	People Tracking
Smart Traffic Count	Queue Analysis

---

Community and Contribution

Community improvements are welcome. If you have a demo, model optimization tip, or pipeline to share, add it under community_projects/ and update community_projects/community_projects.md with usage notes. See the template at community_projects/template_example/ for the expected structure.

---

License

This project is distributed under the MIT License.