CV / CenterPose

CenterPose RegNetX-800MF

reComputer R Series (CM5 + Hailo-10H)

Multi-person pose estimation with 17 COCO keypoints on reComputer R Series with Hailo-8 or Hailo-10H.

81 downloads

Size

13.93 MB

Memory

4GB+

Precision

Hailo HEF / HailoRT

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-hailo10h-centerpose \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --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/centerpose_regnetx_800mf:latest \
  python web_detection.py --model_path model/centerpose_regnetx_800mf.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/centerpose_regnetx_800mf/predict" \
  -F "file=@test.jpg"

Python

import requests

response = requests.post(
    "http://<Board_IP>:8000/api/models/centerpose_regnetx_800mf/predict",
    files={"file": open("test.jpg", "rb")},
    timeout=30,
)
print(response.json())

Model Details

reComputer R Series (CM5 + Hailo-10H)

CenterPose RegNetX-800MF on reComputer R Series (CM5 + Hailo-10H)

CenterPose RegNetX-800MF performs bottom-up, multi-person pose estimation in a single network pass. The Hailo-10H demo returns person boxes and 17 COCO body keypoints, and provides live MJPEG preview, image inference, and offline video analysis through a FastAPI service.

Model information

Property	Value
Architecture	CenterPose with RegNetX-800MF backbone
Task	Multi-person pose estimation
Input	512×512×3 BGR
Output	Person boxes and 17 COCO keypoints
Parameters	12.31M
Hardware AP	43.1 (Hailo Model Zoo reference)
HEF source	Hailo Model Zoo v5.4.0, compiled for Hailo-10H
Runtime	HailoRT 5.1.1

Hardware and host setup

Item	Value
Board	reComputer R Series with Raspberry Pi CM5
Accelerator	Hailo-10H over PCIe, exposed as `/dev/hailo0`
OS	Raspberry Pi OS, aarch64
Validated runtime	HailoRT 5.1.1

bash

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

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

Do not mix HailoRT 4.23.x (Hailo-8) drivers, libraries, or Python wheels into this Hailo-10H container.

Run with the bundled demo video

bash

sudo docker run --rm \
  --name cm5-hailo10h-centerpose \
  --privileged \
  --net=host \
  -e PYTHONUNBUFFERED=1 \
  --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/centerpose_regnetx_800mf:latest \
  python web_detection.py \
    --model_path model/centerpose_regnetx_800mf.hef \
    --video_path video/test.mp4

Open http://<Board_IP>:8000 to view the annotated stream and video-analysis UI.

USB camera mode

bash

sudo docker run --rm --privileged --net=host \
  -e PYTHONUNBUFFERED=1 \
  --device /dev/hailo0:/dev/hailo0 \
  --device /dev/video0:/dev/video0 \
  -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/centerpose_regnetx_800mf:latest \
  python web_detection.py \
    --model_path model/centerpose_regnetx_800mf.hef \
    --camera_id 0

REST API

text

POST http://<Board_IP>:8000/api/models/centerpose_regnetx_800mf/predict

bash

curl -X POST "http://<Board_IP>:8000/api/models/centerpose_regnetx_800mf/predict" \
  -F "file=@test.jpg"

Endpoint	Method	Purpose
`/`	GET	Web preview UI
`/api/models/centerpose_regnetx_800mf/predict`	POST	Single-frame pose inference
`/api/video_feed`	GET	Annotated MJPEG stream
`/api/config`	GET / POST	Read or change confidence/NMS thresholds
`/api/video/upload`	POST	Upload a source video
`/api/video/analyze`	POST	Start offline video analysis
`/api/video/status`	GET	Read analysis progress
`/api/video/download/{filename}`	GET	Download the annotated result

Implementation details

Source: src/hailo10h_centerpose_regnetx_800mf/
Dockerfile: docker/hailo10h/centerpose_regnetx_800mf.dockerfile
HEF: model/centerpose_regnetx_800mf.hef (14,581,760 bytes)
The decoder handles CenterPose heads hm, wh, hps, reg, hm_hp, and hp_offset, then maps detections back through the input letterbox transform.
CenterPose uses the original OpenCV BGR order because normalization is built into this HEF.
The executor uses the HailoRT 5.1.1 create_infer_model API with a persistent configured model and registered FLOAT32 output buffers.

Inputs and Outputs

Input: image, video, or USB camera frame. Output: person boxes, confidences, 17 COCO keypoints, and annotated MJPEG preview.