离线识别中文、日文、韩文和英文,提供 Go、C/C++、Python 与 Android 接口。
| 平台 | 完整版(推荐) |
|---|---|
| Android arm64-v8a / x86_64 | 完整 AAR |
| Linux x64 | 完整运行包 |
| Linux ARM64 | 完整运行包 |
推荐使用包含默认模型和运行库的完整包。Android Core AAR 仅供已有宿主 ORT 的进阶集成,见 Android 指南。Windows/macOS 通过 Go 或 Python 准备完整依赖,不提供平台专属包。
代码接入使用 go get github.com/shiyori/[email protected],详见 Go 指南。命令行接入使用:
go install github.com/shiyori/oneocr-native/cmd/[email protected]
oneocr install
oneocr recognize image.png
oneocr recognize --format json image.pngpython -m pip install "https://github.com/shiyori/oneocr-native/releases/download/v0.1.3/oneocr_native-0.1.3-py3-none-any.whl"
python -m oneocr_native install
python -m oneocr_native recognize image.png
python -m oneocr_native recognize --format json image.pngPaddleOCR 原图完整测试结果:9 张原图的对照图与 JSON
OneOCR v0.1.3 对 testdata 中四张图片的实际输出。仅展示识别置信度 ≥ 0.70 且检测分数 ≥ 0.70 的结果。左图在原图上画框;右图保留原图,在同一位置叠加蓝色识别文字,并加局部浅色底以保持可读性。点击图片可查看高清图。
两张图采用相同裁剪和坐标;外围留白同步裁去,未人工修正识别文字。置信度为未校准模型分数。
显示 6 / 6 条非空识别结果。
| 原图 + 文本框 | 原图 + 蓝色识别文字 |
|---|---|
![]() |
![]() |
显示 47 / 48 条非空识别结果。
| 原图 + 文本框 | 原图 + 蓝色识别文字 |
|---|---|
![]() |
![]() |
显示 84 / 96 条非空识别结果。
| 原图 + 文本框 | 原图 + 蓝色识别文字 |
|---|---|
![]() |
![]() |
显示 96 / 96 条非空识别结果。
| 原图 + 文本框 | 原图 + 蓝色识别文字 |
|---|---|
![]() |
![]() |
下载与安装 · Go 接入 · C 与 C++ 接入 · Python 接入 · Android 接入 · 已有 ONNX Runtime







