纯 C# PP-OCRv6 推理库:多平台手写 Kernel / GEMM 等算子、超高性能、低内存需求、超高准确率。 自带托管 ONNX 解释器,不依赖 Paddle Inference、ONNX Runtime 或 OpenCV 原生库。
TurboOCR, >200 img/s OmnidocBench. TensorRT FP16, PP-OCRv6, HTTP + gRPC
Fast, offline OCR for Node.js & C++. PP-OCRv6 with Core ML / WebGPU hardware acceleration — recognize text in images with confidence scores & coordinates. npm: @arcships/light-ocr
轻量级、无运行时依赖的纯 C PP-OCR 推理运行时,支持 PP-OCRv6 Tiny/Small/Medium、DET/CLS/REC、完整 OCR、SIMD 与 WebAssembly。 Lightweight dependency-free pure-C PP-OCR runtime with PP-OCRv6 Tiny/Small/Medium, full OCR, SIMD and WebAssembly.
PP-OCRv6 纯 Java 图片 OCR(ONNX Runtime,零 PaddlePaddle 依赖,bit-exact 对位 Python,Spring Boot Starter,结构化识别行驶证/身份证/银行卡/驾驶证/营业执照等)
基于 uiautomator2 控件定位的 QQ 宠物自动化工具,支持真机、模拟器,支持多开、后台运行。
CPU-friendly sidecar that OCRs paperless-ngx documents with PaddleOCR PP-OCRv6 and writes text back via the API.