CV / Tiny Yolo

Tiny-YOLOv3

Tiny-YOLOv3 detects 80 COCO object classes on Hailo-8. Unlike the on-chip-NMS models, it outputs raw YOLOv3 heads and does the full grid decode (sigmoid + anchor + NMS) on the CPU. First variant of the tiny_yolo series.

1 ダウンロード

サイズ

8.4 MB

メモリ

4GB+

精度

Hailo HEF / HailoRT 4.23.x

使用しているデバイスを選択してください。セットアップガイドとドキュメントがそれに応じて更新されます。

プラットフォーム

推論エンジン

はじめる

デプロイ

sudo docker run --rm \
  --name cm5-hailo8-tiny-yolov3 \
  --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/tiny_yolov3:latest \
  python web_detection.py --model_path model/tiny_yolov3.hef --video_path video/test.mp4

REST API

REST API を使用して推論を実行します。以下のコマンドをコピーしてください。

Curl

curl -X POST "http://<Board_IP>:8000/api/models/tiny_yolov3/predict" \
  -F "file=@test.jpg"

Python

import requests

response = requests.post(
    "http://<Board_IP>:8000/api/models/tiny_yolov3/predict",
    files={"file": open("test.jpg", "rb")},
    timeout=30,
)
print(response.json())

モデル詳細

Tiny-YOLOv3 on reComputer R Series (CM5 + Hailo-8)

Tiny-YOLOv3 detects 80 COCO object classes on Hailo-8. Unlike the on-chip-NMS models (SSD, EfficientDet, NanoDet), it outputs raw YOLOv3 heads and does the full grid decode (sigmoid + anchor + NMS) on the CPU. First variant of the tiny_yolo series (tiny_yolov3 / tiny_yolov4).

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

Model Info

Property	Value
Architecture	Tiny-YOLOv3 (base/yolo.yaml)
Task	Object detection
Input	416x416x3 RGB (normalize_in_net std=255)
Output	2 raw heads: 13x13x255 (stride 32) + 26x26x255 (stride 16)
Classes	80 (COCO, 0-indexed)
Parameters	8.85M
Operations	5.58G
Postprocess	CPU YOLOv3 decode (NO on-chip NMS)
HEF	Hailo Model Zoo v2.19.0, Hailo-8

Hardware and Host Setup

bash

sudo apt update
sudo apt install hailort hailort-pcie-driver python3-hailort
sudo reboot
hailortcli --version
hailortcli fw-control identify
ls -l /dev/hailo0

Run With Demo Video

bash

sudo docker run --rm \
  --name cm5-hailo8-tiny-yolov3 \
  --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/tiny_yolov3:latest \
  python web_detection.py --model_path model/tiny_yolov3.hef --video_path video/test.mp4

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

USB Camera Mode

bash

sudo docker run --rm \
  --name cm5-hailo8-tiny-yolov3 \
  --privileged \
  --net=host \
  -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/tiny_yolov3:latest \
  python web_detection.py --model_path model/tiny_yolov3.hef --camera_id 0

REST API

text

POST http://<Board_IP>:8000/api/models/tiny_yolov3/predict

bash

curl -X POST "http://<Board_IP>:8000/api/models/tiny_yolov3/predict" \
  -F "file=@test.jpg"

Endpoint	Method	Purpose
`/`	GET	Web preview UI
`/api/models/tiny_yolov3/predict`	POST	Detections (JSON)
`/api/video_feed`	GET	MJPEG preview stream

Implementation Notes

No on-chip NMS — the HEF outputs raw YOLOv3 heads; the app does the full decode on CPU: sigmoid on obj + class scores, anchor-based grid decode, NMS.
Decode (official yolo.py _yolo3_decode): center = (sigmoid(raw_xy) + grid) * stride, scale = exp(raw_wh) * anchor, score = sigmoid(obj) * sigmoid(cls).
normalize_in_net std=255 (÷255) + padding_color=114 (gray). The app feeds raw uint8 RGB pixels after letterboxing.
3 anchors per scale: stride 32 → [[81,82],[135,169],[344,319]], stride 16 → [[23,27],[37,58],[81,82]].
First inference logs head shapes for verification (SOP §10).

Development Notes

Source module: src/rpi5_hailo8_tiny_yolov3/
Dockerfile: docker/hailo8/tiny_yolov3.dockerfile
Container: ghcr.io/seeed-projects/recomputer-hailo8-cv/tiny_yolov3:latest
Family: yolo (variants tiny_yolov3 / tiny_yolov4); this is the v3 build.

入力と出力

Input: image, video, or USB camera frame. Output: COCO 80-class detection boxes with confidences, plus an annotated MJPEG preview.