Files
ssw c303c315e5 rocr: rusto-rs 源码入库(上游 v0.2.5 基线 57609fc + 本地补丁)
- 源码 316 文件收编为 rusto-rs/ 文件夹,随本仓库管理
- 移除内部 .git(补丁已存档 docs/rusto-rs-metal-patch-20260912.patch,上游可随时 re-clone)
- target/ 与 docs/.yarn 缓存不入库;本地改动:ROCR_* 环境变量调优 + Metal 链接修复 + 预处理测试脚本
2026-09-12 17:30:04 +08:00

4.8 KiB

title, description, navigation, seo
title description navigation seo
InitializeConfig Public initialization API reference for model presets, model paths, and auxiliary stages.
icon
i-ph-code-block
title description
InitializeConfig - RustO! API Reference Public initialization API reference for model presets, model paths, and auxiliary stages.

InitializeConfig configures model files, vocabulary dictionaries, and auxiliary networks to instantiate a reusable OCR engine session.

Architecture Contract

flowchart TD
    subgraph Config ["InitializeConfig (Model & Session Setup)"]
        P["Model Preset (ppv6, ppv5, ppv4, ppv3)"]
        D["Detection Model Path (det.mnn)"]
        R["Recognition Model Path (rec.mnn)"]
        V["Vocabulary Dictionary (dict.txt)"]
        C["Optional CLS Classifier (cls.mnn)"]
        O["Optional Page Orientation (orient.mnn)"]
    end

    Config -->|"RustO::initialize(config)"| Engine["RustO Engine Instance<br/>(Reusable Inference Session)"]

Canonical Wire Schema

For C FFI, JNI, and standalone binding JSON serialization:

{
  "template": "ppv6",
  "detection": {
    "modelPath": "models/det.mnn",
    "thresh": 0.3,
    "boxThresh": 0.6,
    "unclipRatio": 2.0,
    "limitSideLen": 736,
    "limitType": "min",
    "useDilation": true
  },
  "recognition": {
    "modelPath": "models/rec.mnn",
    "dictPath": "models/dict.txt"
  },
  "classification": {
    "enabled": true,
    "modelPath": "models/cls.mnn",
    "thresh": 0.9
  },
  "orientation": {
    "enabled": true,
    "modelPath": "models/orient.mnn",
    "thresh": 0.85
  }
}

Presets Reference

Preset Default Detector Resize Default Resize Mode Default Unclip Ratio Compatible Architecture
ppv6 (Default) 736 min 2.0 PP-OCRv6 (Tiny, Small, Medium)
ppv5 736 min 2.0 PP-OCRv5 (Mobile, Server)
ppv4 960 max 1.5 PP-OCRv4 (Mobile, Server)
ppv3 960 max 1.5 PP-OCRv3

Public API Across All Bindings

Rust

use rusto::{InitializeConfig, RustO};

// Method 1: Constructor with default PP-OCRv6 preset
let config = InitializeConfig::ppv6("det.mnn", "rec.mnn", "dict.txt")
    .with_det_thresh(0.3)
    .with_det_box_thresh(0.6)
    .with_unclip_ratio(2.0)
    .with_use_dilation(true)
    .with_cls_threshold("cls.mnn", 0.9)
    .with_orientation_threshold("orient.mnn", 0.85);

let mut ocr = RustO::initialize(config)?;

.NET / C#

using RustODotnet;

var config = new InitializeConfig
{
    Template = "ppv6",
    Detection = new DetectionConfig { ModelPath = "det.mnn" },
    Recognition = new RecognitionConfig { ModelPath = "rec.mnn", DictPath = "dict.txt" },
    Classification = new ClassificationConfig { Enabled = true, ModelPath = "cls.mnn", Thresh = 0.9f },
    Orientation = new OrientationConfig { Enabled = true, ModelPath = "orient.mnn", Thresh = 0.85f }
};

using var ocr = RustO.Initialize(config);

iOS (Swift)

import RustO

let config = InitializeConfig(
    template: "ppv6",
    detection: DetectionConfig(modelPath: "det.mnn"),
    recognition: RecognitionConfig(modelPath: "rec.mnn", dictPath: "dict.txt"),
    classification: ClassificationConfig(enabled: true, modelPath: "cls.mnn", thresh: 0.9),
    orientation: OrientationConfig(enabled: true, modelPath: "orient.mnn", thresh: 0.85)
)

let ocr = try RustO.initialize(config: config)

Android (Kotlin)

import com.byrizki.rusto.*

val config = InitializeConfig(
    template = "ppv6",
    detection = DetectionConfig(modelPath = "det.mnn"),
    recognition = RecognitionConfig(modelPath = "rec.mnn", dictPath = "dict.txt"),
    classification = ClassificationConfig(enabled = true, modelPath = "cls.mnn", thresh = 0.9f),
    orientation = OrientationConfig(enabled = true, modelPath = "orient.mnn", thresh = 0.85f)
)

val ocr = RustO.initialize(context, config)

React Native

import { initialize } from 'react-native-rusto';

await initialize({
  preset: 'ppv6',
  models: {
    detection: 'det.mnn',
    recognition: 'rec.mnn',
    dictionary: 'dict.txt',
    classification: 'cls.mnn',
    orientation: 'orient.mnn',
  },
});

C FFI

#include "librusto.h"

const char *config_json = "{\"template\":\"ppv6\",\"detection\":{\"modelPath\":\"det.mnn\"},\"recognition\":{\"modelPath\":\"rec.mnn\",\"dictPath\":\"dict.txt\"}}";
ROCRHandle *handle = rocr_initialize(config_json);

// Later:
rocr_free(handle);