← Models

CV / YOLOv10

YOLOv10

YOLOv10n and YOLOv10s run with RKNN; YOLOv10n, YOLOv10b, and YOLOv10x are available for Hailo-8.

8 downloads
Size
7.36 MB
Memory
4GB+
Precision
RKNN / Hailo HEF

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

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-yolov10:latest \
  python3 web_detection.py --platform rk3576 --model_path model/yolov10n.rknn --video_path video/test.mp4

Model Details

reComputer RK

YOLOv10 on reComputer RK3576 and RK3588

This RKNN deployment comes from the YOLOv10 modules in reComputer-RK-CV. It implements the model's end-to-end, NMS-free two-stage Top-K post-processing.

Model information

PropertyValue
TaskCOCO 80-class object detection
Input640 x 640, letterboxed and converted from BGR to RGB
Variantsyolov10n.rknn, yolov10s.rknn
Defaultmodel/yolov10n.rknn
Post-processingEnd-to-end, NMS-free Top-K selection

Run the service

RK3576

bash
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-yolov10:latest \
  python3 web_detection.py --platform rk3576 \
  --model_path model/yolov10n.rknn --video_path video/test.mp4

RK3588

bash
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/rk3588-yolov10:latest \
  python3 web_detection.py --platform rk3588 \
  --model_path model/yolov10n.rknn --video_path video/test.mp4

Select YOLOv10s with --model_path model/yolov10s.rknn. Replace the video argument with --camera_id -1 for upload-only mode.

Startup arguments

ArgumentDefaultDescription
--platformRequiredrk3576 or rk3588.
--model_pathRequiredYOLOv10n or YOLOv10s RKNN file.
--camera_id1Camera index; -1 disables capture.
--video_pathNoneLooping local MP4; overrides the camera.
--class_pathCOCO labelsOptional replacement class list.
--host / --port0.0.0.0 / 8000Service address and port.

REST API

Endpoint: POST /api/models/yolov10/predict

bash
curl -X POST "http://<BOARD_IP>:8000/api/models/yolov10/predict" \
  -F "file=@bus.jpg" -F "conf=0.25"

Example response:

json
{
  "success": true,
  "source": "uploaded image",
  "predictions": [
    {
      "class": "bus",
      "confidence": 0.91,
      "box": {"x1": 100, "y1": 120, "x2": 520, "y2": 430}
    }
  ],
  "image": {"width": 640, "height": 480}
}

The API accepts iou for compatibility with the standard detection service, but YOLOv10's NMS-free post-process does not use it. The global confidence setting from GET/POST /api/config applies to preview and offline analysis.

The service also exposes health, MJPEG, video upload, asynchronous analysis, status, list, and download endpoints. Runtime files are written under workspace/ by default. A replacement model must retain the input size and output tensors expected by the NMS-free decoder in web_detection.py.

Build locally

bash
docker build -f docker/rk3576/yolov10.dockerfile \
  -t rk3576-yolov10:local src/rk3576_yolov10

docker build -f docker/rk3588/yolov10.dockerfile \
  -t rk3588-yolov10:local src/rk3588_yolov10

Inputs and Outputs

Input: image, video, or USB camera frame. Output: detection boxes, classes, confidences, and annotated MJPEG preview.