CV / YOLO11
YOLO11
reComputer R Series (CM5 + Hailo-10H)
Lichtgewicht YOLO11 nano model voor snelle edge-deployment.
67 downloads
Grootte
~5MBGeheugen
1GB+Precisie
FP16Kies het apparaat dat je gebruikt. De installatiehandleiding en documentatie worden dienovereenkomstig bijgewerkt.
Voorbeeld
Aan de slag
Implementeren
sudo docker run --rm \
--name cm5-hailo10h-yolo11n \
--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/yolov11n:latest \
python web_detection.py --model_path model/yolov11n.hef --video_path video/test.mp4Modeldetails
reComputer CM5 + Hailo-10HYOLOv11 op reComputer CM5 + Hailo-10H
YOLOv11 (varianten n/s/m/l/x) voert COCO-objectdetectie met 80 klassen uit op de Hailo-10H via HailoRT 5.1.1. De NMS draait on-chip (HPP), dus de applicatie parseert de post-NMS 80x5x100 tensor rechtstreeks.
Modelinformatie
| Variant | Invoer | Parameters | Operaties | Hailo-10H HEF |
|---|---|---|---|---|
| YOLOv11n | 640x640x3 | 2.6M | 6.55G | 5.3 MB |
| YOLOv11s | 640x640x3 | 9.4M | 21.5G | 16.3 MB |
| YOLOv11m | 640x640x3 | 20.1M | 68.0G | 27.0 MB |
| YOLOv11l | 640x640x3 | 25.3M | 86.9G | 27.9 MB |
| YOLOv11x | 640x640x3 | 56.9M | 194G | 54.2 MB |
| Eigenschap | Waarde |
|---|---|
| Taak | Objectdetectie |
| Uitvoer | On-chip NMS-tensor 80x5x100 (rijen: ymin,xmin,ymax,xmax,score, genormaliseerd) |
| Klassen | 80 (COCO, indexvolgorde van de Model Zoo) |
| HEF | Hailo Model Zoo v5.4.0, Hailo-10H |
Hardware- en hostconfiguratie
| Onderdeel | Waarde |
|---|---|
| Board | reComputer R Series met Raspberry Pi CM5 |
| Accelerator | Hailo-10H via PCIe, beschikbaar als /dev/hailo0 |
| Hoststuurprogramma | hailo-h10-all apt-pakket |
| Runtime | HailoRT 5.1.1 |
| Python in de container | 3.13, aarch64 |
bash
sudo apt update
sudo apt install hailo-h10-all -y
sudo reboot
# After reboot
hailortcli fw-control identify
ls /dev/hailo0Uitvoeren met demovideo
Stap A: Images ophalen
bash
sudo docker pull ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov11n:latest
sudo docker pull ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov11s:latest
sudo docker pull ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov11m:latest
sudo docker pull ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov11l:latest
sudo docker pull ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov11x:latestStap B: Container starten
Kies de opdracht die hoort bij de modelvariant.
YOLOv11n
bash
sudo docker run --rm \
--name cm5-hailo10h-yolov11n \
--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/yolov11n:latest \
python web_detection.py --model_path model/yolov11n.hef --video_path video/test.mp4YOLOv11s
bash
sudo docker run --rm --name cm5-hailo10h-yolov11s --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/yolov11s:latest python web_detection.py --model_path model/yolov11s.hef --video_path video/test.mp4YOLOv11m
bash
sudo docker run --rm --name cm5-hailo10h-yolov11m --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/yolov11m:latest python web_detection.py --model_path model/yolov11m.hef --video_path video/test.mp4YOLOv11l
bash
sudo docker run --rm --name cm5-hailo10h-yolov11l --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/yolov11l:latest python web_detection.py --model_path model/yolov11l.hef --video_path video/test.mp4YOLOv11x
bash
sudo docker run --rm --name cm5-hailo10h-yolov11x --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/yolov11x:latest python web_detection.py --model_path model/yolov11x.hef --video_path video/test.mp4Open http://<Board_IP>:8000 om de webpreview te bekijken.
REST API
Elke variant heeft een eigen voorspelendpoint:
| Variant | Endpoint |
|---|---|
| YOLOv11n | POST /api/models/yolov11n/predict |
| YOLOv11s | POST /api/models/yolov11s/predict |
| YOLOv11m | POST /api/models/yolov11m/predict |
| YOLOv11l | POST /api/models/yolov11l/predict |
| YOLOv11x | POST /api/models/yolov11x/predict |
Voorbeeld:
bash
curl -X POST "http://<Board_IP>:8000/api/models/yolov11n/predict" -F "file=@test.jpg"Implementatienotities
- On-chip NMS (HPP,
meta_arch=yolov8overgeërfd van de yolov11n-configuratie); de applicatie parseert de post-NMS 80x5x100 tensor — rijen [ymin, xmin, ymax, xmax, score], genormaliseerd ten opzichte van de 640x640 invoer. - Voorverwerking: letterbox (grijs 114) + RGB, rauwe uint8 (normalisatie 0/255).
- De executor is de HailoRT 5.1.1 multi-uitvoer FLOAT32-wrapper; de eerste inferentie print het uitvoerbereik volgens de SOP-acceptatiechecklist.
Ontwikkelingsnotities
- Bronmodules:
src/hailo10h_yolov11{n,s,m,l,x}/ - Dockerfiles:
docker/hailo10h/yolov11{n,s,m,l,x}.dockerfile - Containers:
ghcr.io/seeed-projects/recomputer-hailo10h-cv/yolov11{n,s,m,l,x}:latest - HEF's: hailo10h-builds van Hailo Model Zoo v5.4.0.