reComputer-Hailo10H-CV

[English] | 中文

This project provides industrial-grade Computer Vision (CV) solutions for the reComputer R Series (Raspberry Pi 5) with Hailo-10H accelerator. Models are pre-compiled from the Hailo Model Zoo and packaged with Docker for one-click deployment.

Project Architecture

Each model is self-contained with its own inference code, model file, and Docker configuration:

reComputer-Hailo10H-CV/
├── .github/workflows/       # CI/CD auto-build Docker images
├── tools/                   # Scaffolding scripts
├── docs/                    # Documentation
├── docker/hailo10h/         # Dockerfiles per model
├── src/
│   └── hailo10h_<model>/
│       ├── web_detection.py # FastAPI inference service
│       ├── py_utils/        # HailoInfer + utilities
│       ├── hailort-packages/ # HailoRT Python wheel
│       ├── model/           # HEF model file
│       ├── video/           # Demo video
│       └── requirements.txt

Supported Platforms

Platform	Chip	Accelerator	Computing Power
reComputer R Series	Raspberry Pi 5	Hailo-10H	10 TOPS

Available Models

Model	Task	Params	Input	Metric	HEF
STDC1	Semantic Segmentation	8.27M	1024×1920	73.7% mIoU	Download
FCN8 ResNet-18	Semantic Segmentation	11.20M	1024×1920	69.2% mIoU	Download
MSPN RegNetX-800MF	Pose Estimation	7.17M	256×192	69.8% AP	Download
LightFace Slim	Face Detection	0.26M	240×320	39.3% mAP	Download
Face Landmarks Lite	Facial Landmarks	0.6M	192×192	—	Source
ArcFace MobileFaceNet	Face Embedding	2.04M	112×112	99.4% LFW	Source
YOLO26n-seg	Instance Segmentation	2.7M	640×640	—	Source
YOLO26s-seg	Instance Segmentation	10.4M	640×640	—	Source
YOLO26m-seg	Instance Segmentation	23.6M	640×640	—	Source
YOLOv8n-seg	Instance Segmentation	3.4M	640×640	—	Source
YOLOv8s-seg	Instance Segmentation	11.8M	640×640	—	Source
YOLOv8m-seg	Instance Segmentation	27.3M	640×640	—	Source

Quick Start

1. Install Docker

curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh --mirror Aliyun
sudo systemctl enable docker && sudo systemctl start docker

2. Install HailoRT (Hailo-10H)

The containers use Python 3.13 and the HailoRT 5.1.1 Python wheel. Keep the host driver, runtime library, and firmware on the matching HailoRT release.

sudo apt update
sudo apt install hailo-h10-all -y
sudo reboot

# Verify
hailortcli fw-control identify
ls /dev/hailo0

3. Pull and Run

# Pull from GHCR
docker pull ghcr.io/seeed-projects/recomputer-hailo10h-cv/stdc1:latest

# Run
sudo docker run --rm --privileged --net=host \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.5.1.1:/usr/lib/libhailort.so.5.1.1:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo10h-cv/stdc1:latest

Open http://<device_ip>:8000 to view the web preview.

Or Build Locally

cd src/hailo10h_stdc1
docker build -f ../../docker/hailo10h/stdc1.dockerfile -t stdc1 .

REST API

Prediction

curl -X POST http://<device_ip>:8000/api/models/stdc1/predict \
  -F "file=@test.jpg"

The web preview is available at /, and the MJPEG stream is available at /api/video_feed. Model-specific endpoints are documented in each module README.

Model Sources

All HEF models are from Hailo Model Zoo, pre-compiled for Hailo-10H.

Contributing

1. Use scaffold: python tools/scaffold_hailo_model.py <model_name> --task <task>
2. Add HEF to model/
3. Update web_detection.py post-processing
4. Add HailoRT wheel to hailort-packages/
5. Add model to recomputer-ai-lab
6. See docs/HAILO_MODEL_PORTING_zh.md for details

License

Open source. See individual model directories for license information.