CV / YOLOv6

YOLOv6

YOLOv6 COCO object detection on reComputer RK3576/RK3588 through RKNN and on reComputer R Series with Hailo-8 or Hailo-10H through HailoRT.

11 downloads

Precision

RKNN / Hailo HEF

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

Platform

Size

Inference Engine

Getting Started

Deploy

sudo docker run --rm --privileged --net=host --device /dev/dri/renderD128:/dev/dri/renderD128 \
  -v /proc/device-tree/compatible:/proc/device-tree/compatible:ro \
  ghcr.io/seeed-projects/recomputer-rk-cv/rk3576-yolov6:latest \
  python3 web_detection.py --platform rk3576 --model_path model/yolov6n.rknn --video_path video/test.mp4

REST API

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

Curl

curl http://localhost:8080/v1/chat/completions -d '{
  "model": "yolov6",
  "messages": [{"role": "user", "content": "Hello"}]
}'

Python

import requests

resp = requests.post(
    "http://localhost:8080/v1/chat/completions",
    json={"model": "yolov6", "messages": [{"role": "user", "content": "Hello"}]},
)
print(resp.json())

Model Details

reComputer R Series (CM5 + Hailo-8)

YOLOv6n on reComputer R Series (CM5 + Hailo-8)

YOLOv6n performs COCO 80-class object detection on Hailo-8 through HailoRT. The compiled HEF either returns the on-chip HPP NMS result or the nine raw split heads; the app handles both and logs which layout the HEF produced.

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

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
Python in container	3.11, aarch64

Install the Hailo-8 packages on the host and confirm that the driver and device are ready before starting the container:

bash

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

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

The expected HailoRT version for this image is 4.23.x.

Run With Demo Video

bash

sudo docker run --rm \
  --name cm5-hailo8-yolov6n \
  --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/yolov6n:latest \
  python web_detection.py --model_path model/yolov6n.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-yolov6n \
  --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/yolov6n:latest \
  python web_detection.py --model_path model/yolov6n.hef --camera_id 0

HEF File

HEF	Size	Notes
`yolov6n.hef`	5,774,047 (5.5 MB)	The HEF comes from the Hailo Model Zoo v2.19.0 build for Hailo-8; parameters 4.32M, operations 11.12G.

Source: https://hailo-model-zoo.s3.eu-west-2.amazonaws.com/ModelZoo/Compiled/v2.19.0/hailo8/yolov6n.hef.

REST API

text

POST http://<Board_IP>:8000/api/models/yolov6n/predict

bash

curl -X POST "http://<Board_IP>:8000/api/models/yolov6n/predict" \
  -F "file=@test.jpg"

Endpoint	Method	Purpose
`/`	GET	Web preview UI
`/api/models/yolov6n/predict`	POST	Detections (JSON)
`/api/video_feed`	GET	MJPEG preview stream

Implementation Notes

The Model Zoo base/yolov6.yaml uses hpp=true, meta_arch=yolo_v6, score_threshold=0.03 and nms_iou_thresh=0.65; info.output_shape documents the nine raw split heads (4 box, 1 objectness, 80 classes per stride 32/16/8).
The module handles both compile layouts: when the HEF already ran NMS (HPP) it parses the post-NMS tensor (compact per-class buffer, dense Cx5xD / CxDx5, ragged NMS-by-score list); when the HEF returns the raw split heads it decodes the distances around the cell centre (stride units), multiplies sigmoid(classes) by sigmoid(objectness) and runs a per-class NMS with the IOU slider value.
[YOLOv6n] layout=... is printed once on the first inference so the layout can be confirmed on hardware.
The HEF bakes normalize_in_net (mean 0 / std 255) and padding_color=114: the app letterboxes with gray (114) and feeds raw uint8 RGB pixels — no manual normalization.
Class IDs 0..79 index the standard COCO class list directly (the Model Zoo evaluation uses labels_offset=1).
Defaults: confidence 0.25; with the on-chip NMS layout the IOU slider has no effect, with the raw-head layout it drives the host-side NMS.

Development Notes

Source module: src/hailo10h_yolov6n/ (Hailo-10H) or src/rpi5_hailo8_yolov6n/ (Hailo-8) in the matching model repository
Container: ghcr.io/seeed-projects/recomputer-hailo8-cv/yolov6n:latest
Only YOLOv6n is published by the Hailo Model Zoo; YOLOv6s/m/l have no Hailo HEF.
Model licence: GPL-3.0 (upstream meituan/YOLOv6).

Inputs and Outputs

Input: image, video frame, or camera frame. Output: COCO classes, confidence scores, and boxes.