CV / LightFace

LightFace Slim

Ultra-lightweight face detection (0.26M params) on reComputer R Series with Hailo-8 or Hailo-10H.

64 downloads

Memory

4GB+

Precision

Hailo HEF

Choose the device you're using, the set up guide and documentation will update accordingly.

Platform

Inference Engine

Getting Started

Deploy

sudo docker run --rm \
  --name cm5-hailo8-lightface \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.4.23.0:/usr/lib/libhailort.so.4.23.0:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo8-cv/lightface_slim:latest \
  python web_detection.py --model_path model/lightface_slim.hef --video_path video/test.mp4

REST API

Use the REST API to run inference. Copy the commands below.

Curl

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

Python

import requests

resp = requests.post(
    "http://<Board_IP>:8000/api/models/lightface_slim/predict",
    files={"file": open("test.jpg", "rb")},
    timeout=30,
)
print(resp.json())

Model Details

reComputer R Series (CM5 + Hailo-8)

LightFace Slim on reComputer R Series (CM5 + Hailo-8)

Ultra-lightweight face detection (0.26M params) running on Hailo-8 via HailoRT 4.23. The HEF exposes raw SSD-style heads (no on-chip NMS): the host builds the 4,420 anchors, runs the SSD box decode (variances 10/5), the softmax over the background/face logits, the score cut and greedy NMS.

This page targets reComputer R Series (CM5 + Hailo-8) with a PCIe Hailo-8 accelerator.

Model Info

Property	Value
Architecture	Ultra-Light-Fast-Generic-Face-Detector-1MB (SSD-style, strides 8/16/32/64)
Task	Face Detection (single "face" class, WIDER FACE)
Input	240×320×3 (the app feeds uint8 BGR; mean 127 / std 128 are baked into the HEF)
Output	8 raw heads: 4 scales × {bbox deltas, background + face logits}
Decode	Host-side SSD decode (variances 10/5) + softmax + greedy NMS (`meta_arch=retinaface`)
Parameters	0.26M
Operations	0.16G
Reference mAP	39.71 (WIDER FACE, full precision, Hailo Model Zoo; not measured on this board)
License	MIT (Ultra-Light-Fast-Generic-Face-Detector-1MB)
Source	Hailo Model Zoo

Hardware and Host Setup

Item	Value
Board	reComputer R Series with Raspberry Pi CM5
Accelerator	Hailo-8 over PCIe, exposed as `/dev/hailo0`
Runtime	HailoRT 4.23.x (host / container must share major.minor)
Python in container	3.11, aarch64

bash

sudo apt update
sudo apt install hailort hailort-pcie-driver python3-hailort
sudo reboot

# After reboot
hailortcli fw-control identify
ls /dev/hailo0

# 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

Install hailort hailort-pcie-driver python3-hailort directly — not hailo-all, which can pull the Hailo-10H / HailoRT 5.x packages.

Run With Demo Video

bash

sudo docker run --rm --privileged --net=host \
  --name cm5-hailo8-lightface \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  -v /usr/lib/libhailort.so.4.23.0:/usr/lib/libhailort.so.4.23.0:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo8-cv/lightface_slim:latest \
  python web_detection.py --model_path model/lightface_slim.hef --video_path video/test.mp4

USB Camera Mode

bash

sudo docker run --rm --privileged --net=host \
  --name cm5-hailo8-lightface \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  --device /dev/video0:/dev/video0 \
  -v /usr/lib/libhailort.so.4.23.0:/usr/lib/libhailort.so.4.23.0:ro \
  -v /usr/lib/libhailort.so:/usr/lib/libhailort.so:ro \
  ghcr.io/seeed-projects/recomputer-hailo8-cv/lightface_slim:latest \
  python web_detection.py --model_path model/lightface_slim.hef --camera_id 0

Open http://<Board_IP>:8000 for the Web preview.

HEF Files

HEF	Size	Notes
`lightface_slim.hef`	1,907,332 bytes	Hailo Model Zoo v2.19.0 Hailo-8 build; 8 raw heads

Source:

text

https://hailo-model-zoo.s3.eu-west-2.amazonaws.com/ModelZoo/Compiled/v2.19.0/hailo8/lightface_slim.hef

SHA256: e448c0f692f9e907bb7328fc2305ced4d3569f1e0aa26d44439d1a6f003e9314

REST API

text

POST http://<Board_IP>:8000/api/models/lightface_slim/predict

bash

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

Response (box is [x1, y1, x2, y2] in original-frame pixels):

json

{
  "predictions": [
    {"box": [220, 156, 260, 196], "confidence": 0.98}
  ]
}

Endpoint	Method	Purpose
`/`	GET	Web preview UI
`/api/models/lightface_slim/predict`	POST	Face boxes (JSON)
`/api/video_feed`	GET	MJPEG preview stream
`/api/config`	GET / POST	Confidence / IoU thresholds

Implementation Notes

Anchor set from the zoo config: steps 8/16/32/64, min_sizes [10,16,24] / [32,48] / [64,96] / [128,192,256] → 4,420 anchors.
Preview confidence cut 0.5; the zoo evaluation uses 0.1 with NMS IoU 0.3.
The frame is resized with the aspect ratio preserved and padded bottom/right with 0, matching the zoo's face-detection preprocessing.

Development

Source: src/rpi5_hailo8_lightface_slim/
Dockerfile: docker/hailo8/lightface_slim.dockerfile
Validation log: src/rpi5_hailo8_lightface_slim/TEST_REPORT.md

Inputs and Outputs

Input: image (240x320 RGB). Output: face bounding boxes with confidence scores.