CV / YOLOX

YOLOX

Select an RKNN YOLOX-S/M model or a Hailo-8 Tiny/S-Leaky/L-Leaky HEF for the target platform.

5 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-yolox:latest \
  python3 web_detection.py --platform rk3576 --model_path model/yolox_s.rknn \
  --class_path model/coco_80_labels_list.txt --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": "yolox-rknn",
  "messages": [{"role": "user", "content": "Hello"}]
}'

Python

import requests

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

Model Details

reComputer RK

YOLOX on reComputer RK3576 and RK3588

This page is based on the YOLOX services in reComputer-RK-CV. The runtime performs RKNN inference, YOLOX branch decoding, objectness/class-score fusion, and class-aware NMS.

Model information

Property	Value
Task	COCO 80-class object detection
Input	`640 x 640` letterboxed RGB
Variants	`yolox_s.rknn`, `yolox_m.rknn`
Default	`model/yolox_s.rknn`
Labels	`model/coco_80_labels_list.txt`

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-yolox:latest \
  python3 web_detection.py --platform rk3576 \
  --model_path model/yolox_s.rknn \
  --class_path model/coco_80_labels_list.txt --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-yolox:latest \
  python3 web_detection.py --platform rk3588 \
  --model_path model/yolox_s.rknn \
  --class_path model/coco_80_labels_list.txt --video_path video/test.mp4

Use model/yolox_m.rknn for the medium variant. Keep the class-path argument when switching models. Open http://<BOARD_IP>:8000 or /docs.

Startup arguments

Argument	Default	Description
`--platform`	Required	`rk3576` or `rk3588`.
`--model_path`	Required	YOLOX-S or YOLOX-M RKNN file.
`--class_path`	Required by the packaged command	COCO class-name file.
`--camera_id`	`1`	Camera index; `-1` enables uploads only.
`--video_path`	None	Looping local MP4; overrides the camera.
`--host` / `--port`	`0.0.0.0` / `8000`	Service address and port.

REST API

Endpoint: POST /api/models/yolox/predict

bash

curl -X POST "http://<BOARD_IP>:8000/api/models/yolox/predict" \
  -F "file=@bus.jpg" -F "conf=0.25" -F "iou=0.45"

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 service accepts image, uploaded-video frame, and current-source requests. GET/POST /api/config controls obj_thresh and nms_thresh; request-level conf and iou override them. Health, MJPEG, video upload, asynchronous analysis, status, list, and download endpoints are also available.

Input is letterboxed, converted to RGB, decoded by branch, and processed with objectness/class-score fusion and class-aware NMS. Replacement models must match this included YOLOX decoder and class configuration.

Build locally

bash

docker build -f docker/rk3576/yolox.dockerfile \
  -t rk3576-yolox:local src/rk3576_yolox

docker build -f docker/rk3588/yolox.dockerfile \
  -t rk3588-yolox:local src/rk3588_yolox

Inputs and Outputs

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