README-en.md

🚀 vision-simple 🚀

english | 简体中文

vision-simple is a cross-platform visual inference library based on C++23, designed to provide out-of-the-box inference capabilities. With Docker, users can quickly set up inference services. This library currently supports popular YOLO models (including YOLOv10 and YOLOv11) and some OCR models (such as PaddleOCR). It features a built-in HTTP API, making the service more accessible. Additionally, vision-simple uses the ONNXRuntime engine, which supports multiple Execution Providers such as DirectML, CUDA, TensorRT, and can be compatible with specific hardware devices (such as RockChip's RKNPU), offering more efficient inference performance.

🚀 Features

• Cross-platform: Supports windows/x64, linux/x86_64, linux/arm64,and linux/riscv64
• Multi-device: Supports CPU, GPU, and RKNPU
• Small size: The statically compiled version is under 20 MiB, with YOLO and OCR inference occupying 300 MiB of memory
• Fast deployment:
  • One-click compilation: Provides verified build scripts for multiple platforms
  • Container deployment: One-click deployment with docker, podman, or containerd
  • HTTP Service: Offers a HTTP API for non-real-time applications

yolov11n 3440x1440@60fps+

OCR (HTTP API)

🚀 Using vision-simple

Deploy HTTP Service with docker

1. Start the server project:

docker run -it --rm --name vs -p 11451:11451 lonacn/vision_simple:0.4.1-cpu-x86_64

2. Open the Swagger online editor and allow the site’s unsafe content.
3. Copy the content from doc/openapi/server.yaml into the Swagger editor.
4. On the right panel of the editor, select the APIs you want to test

🚀 Quick Start for Development

### Build Project
#### windows/x64
- xmake >= 2.9.7
- msvc with C++23
- Windows 11
```powershell
# pull project
git clone https://github.com/lona-cn/vision-simple.git
cd vision-simple
# setup sln
./scripts/dev-vs.bat
# run server
xmake build server
xmake run server

linux/x86_64

• xmake >= 2.9.7
• gcc-13
• Debian 12 / Ubuntu 2022

# pull project
git clone https://github.com/lona-cn/vision-simple.git
cd vision-simple
# build release
./scripts/build-release.sh
# run server
xmake build server
xmake run server

Docker Image

All Dockerfiles are located in the docker/ directory.

# pull project
git clone https://github.com/lona-cn/vision-simple.git
cd vision-simple
# Build the project
docker build -t vision-simple:latest -f  docker/Dockerfile.debian-bookworm-x86_64-cpu .
# Run the container, the default configuration will use CPU inference and listen on port 11451
docker run -it --rm -p 11451:11451 --name vs vision-simple

dev YOLOv11 Inference with vision-simple

#include <Infer.h>
#include <opencv2/opencv.hpp>
using namespace vision_simple;
template <typename T>
struct DataBuffer
{
    std::unique_ptr<T[]> data;
    size_t size;
    std::span<T> span(){return std::span{data.get(), size};}
};

extern std::expected<DataBuffer<uint8_t>, InferError> ReadAll(const std::string& path);

int main(int argc,char *argv[]){
    auto data = ReadAll("assets/hd2-yolo11n-fp32.onnx");
    auto image = cv::imread("assets/hd2.png");
    auto ctx = InferContext::Create(InferFramework::kONNXRUNTIME, InferEP::kDML);
    auto infer_yolo = InferYOLO::Create(**ctx, data->span(), YOLOVersion::kV11);
    auto result = infer_yolo->get()->Run(image, 0.625);
    // do what u want
    return 0;
}

<div align="center">📄 License</div>
The copyrights for the YOLO models and PaddleOCR models in this project belong to the original authors.

This project is licensed under the Apache-2.0 license.