v0.2 to v0.3

RustO! v0.3.0 is a major architectural release that transitions the entire core engine to pure Rust, eliminates C++/CMake toolchain dependencies, and streamlines the configuration API across all supported languages.

⚡ Quick Upgrade Summary

1. Engine Architecture & Model Formats

Zero C++ / CMake Toolchain

In v0.2, RustO relied on Alibaba MNN via C FFI bindings. This required CMake, platform-specific C++ compilers, and complex NDK/Xcode toolchain integration.

In v0.3, the default inference engine is 100% pure Rust powered by RTen. Compiling RustO only requires standard Cargo (cargo build), with zero native shared libraries or build-time dependencies.

Transitioning Model Formats

Proprietary .mnn models are deprecated. Update your model references to .rten (FlatBuffers) or .onnx:

- det_model: "models/det.mnn"
- rec_model: "models/rec.mnn"
+ det_model: "models/det.rten"
+ rec_model: "models/rec.rten"

Pre-converted .rten and .onnx models are available on GitHub Releases, npm, and NuGet.

2. Session API: EngineSession

The concrete MnnSession struct has been removed in favor of EngineSession, which implements the generic InferenceSession trait.

Before (v0.2)

use rusto::session::MnnSession;

let det_session = MnnSession::from_file("models/det.mnn")?;

After (v0.3)

use rusto::session::EngineSession;

// Using built-in RTen backend:
let det_session = EngineSession::from_det_config(&config.det)?;
let rec_session = EngineSession::from_rec_config(&config.rec)?;

// Or wrap custom third-party engines (e.g. ONNX Runtime ort, Tract):
let custom_session = EngineSession::from_custom(my_session_box);
let ocr = RustO::with_custom_engines(config, det_session, rec_session)?;

3. Configuration & Runtime Options Refactoring

In v0.2, image preprocessing thresholds and post-processing flags were coupled inside static configuration structures. In v0.3, options are decoupled into two distinct categories:

1. InitializeConfig: One-time static setup for model paths, vocabulary dictionary, and thread allocation.
2. OcrRunOptions: Per-request options passed into detect_text(...) or run_ocr(...).

Flattened Options Structure

The legacy nested preprocessing block has been removed. Detection thresholds and post-processing limits are now direct sibling fields on OcrRunOptions:

Before (v0.2)

// v0.2: Runtime thresholds mutated static engine configuration
let mut config = Config::from_files("det.mnn", "rec.mnn", "dict.txt")?;
config.preprocessing.det_db_thresh = 0.3;
config.preprocessing.det_db_box_thresh = 0.6;
config.preprocessing.det_db_unclip_ratio = 1.6;

let mut ocr = RustO::new(config)?;
let result = ocr.run(&image)?;

After (v0.3)

use rusto::{InitializeConfig, OcrRunOptions, OutputGranularity, RustO};

// 1. Static initialization (thread-safe, load once)
let mut ocr = RustO::initialize(InitializeConfig::ppv6(
    "models/det.rten",
    "models/rec.rten",
    "models/dict.txt",
))?;

// 2. Per-request options (stateless, concurrent)
let mut options = OcrRunOptions::default();
options.detection.box_threshold = 0.6;
options.postprocess.unclip_ratio = 1.6;
options.output_granularity = OutputGranularity::Lines;

// Optional: Document calibration & resolution scaling
options.calibration.descreen = true;
options.optimization.max_side_len = 960;

let result = ocr.detect_text(&image, &options)?;

4. Package Namespaces Migration (@rustors/*)

All npm packages have migrated to the official @rustors organization scope:

Updating package.json

  "dependencies": {
-   "react-native-rusto": "^0.2.4",
-   "rusto-models-ppocrv6-tiny": "^0.2.4"
+   "@rustors/react-native": "^0.3.0",
+   "@rustors/model-ppocrv6-tiny": "^0.3.0"
  }

5. Mobile & Native Packages

• Android (JitPack / Gradle):

  // app/build.gradle
  implementation 'com.github.byrizki.rusto-rs:rusto-android:0.3.0'
  implementation 'com.github.byrizki.rusto-rs:rusto-models-ppocrv6-tiny:0.3.0'
  

• iOS (CocoaPods):

  # Podfile
  pod 'RustO', '~> 0.3.0'
  pod 'RustO-Models-PPOCRv6-Tiny', '~> 0.3.0'
  

• .NET (NuGet):

  dotnet add package RustODotnet --version 0.3.0
  dotnet add package RustODotnet.Models.PPOCRv6.Tiny --version 0.3.0
  

6. Migration Checklist

• Step 1: Replace model files (.mnn) with .rten or .onnx models.
• Step 2: Update dependencies to 0.3.0 (using @rustors/* for npm).
• Step 3: Replace MnnSession references with EngineSession.
• Step 4: Separate static InitializeConfig from per-image OcrRunOptions.
• Step 5: Flatten configuration fields (replace nested preprocessing.* with options.detection.* and options.postprocess.*).
• Step 6: Validate OCR output and bounding boxes on test images.