Crate ocr_rs 

Rust PaddleOCR

A high-performance OCR library based on PaddleOCR models, using the MNN inference framework.

Version 2.0 New Features

• New API Design: Complete layered API from low-level models to high-level pipeline
• Flexible Model Loading: Support loading models from file paths or memory bytes
• Configurable Detection Parameters: Support custom detection thresholds, resolution, etc.
• GPU Acceleration: Support multiple GPU backends including Metal, OpenCL, Vulkan
• Batch Processing: Support batch text recognition for improved throughput

Quick Start

Simple Usage - Using High-Level API (Recommended)

use ocr_rs::{OcrEngine, OcrEngineConfig};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create OCR engine
    let engine = OcrEngine::new(
        "models/det_model.mnn",
        "models/rec_model.mnn",
        "models/ppocr_keys.txt",
        None, // Use default config
    )?;

    // Open and recognize image
    let image = image::open("test.jpg")?;
    let results = engine.recognize(&image)?;

    for result in results {
        println!("Text: {}, Confidence: {:.2}%", result.text, result.confidence * 100.0);
        println!("Position: ({}, {})", result.bbox.rect.left(), result.bbox.rect.top());
    }

    Ok(())
}

Advanced Usage - Using Low-Level API

use ocr_rs::{DetModel, RecModel, DetOptions, DetPrecisionMode};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create detection model
    let det = DetModel::from_file("models/det_model.mnn", None)?
        .with_options(DetOptions::fast());

    // Create recognition model
    let rec = RecModel::from_file("models/rec_model.mnn", "models/ppocr_keys.txt", None)?;

    // Load image
    let image = image::open("test.jpg")?;

    // Detect and crop text regions
    let detections = det.detect_and_crop(&image)?;

    // Recognize each text region
    for (cropped_img, bbox) in detections {
        let result = rec.recognize(&cropped_img)?;
        println!("Position: ({}, {}), Text: {}",
            bbox.rect.left(), bbox.rect.top(), result.text);
    }

    Ok(())
}

GPU Acceleration

use ocr_rs::{OcrEngine, OcrEngineConfig, Backend};

let config = OcrEngineConfig::new()
    .with_backend(Backend::Metal);  // macOS/iOS
    // .with_backend(Backend::OpenCL);  // Cross-platform

let engine = OcrEngine::new(det_path, rec_path, charset_path, Some(config))?;

Module Structure

• mnn: MNN inference engine wrapper, provides low-level inference capabilities
• det: Text detection module (DetModel), detects text regions in images
• rec: Text recognition module (RecModel), recognizes text content
• engine: High-level OCR pipeline (OcrEngine), all-in-one OCR solution
• preprocess: Image preprocessing utilities, including normalization, scaling, etc.
• postprocess: Post-processing utilities, including NMS, box merging, sorting, etc.
• error: Error types OcrError

API Hierarchy

┌─────────────────────────────────────────┐
│        OcrEngine (High-Level API)       │
│   Complete detection and recognition    │
├─────────────────────────────────────────┤
│     DetModel      │      RecModel       │
│  Detection Model  │  Recognition Model  │
├─────────────────────────────────────────┤
│          InferenceEngine (MNN)          │
│         Low-level inference engine      │
└─────────────────────────────────────────┘

Supported Models

• PP-OCRv4: Stable version, good compatibility
• PP-OCRv5: Recommended version, supports multiple languages, higher accuracy
• PP-OCRv5 FP16: Efficient version, faster inference, lower memory usage

Re-exports

pub use det::DetModel;

pub use det::DetOptions;

pub use det::DetPrecisionMode;

pub use engine::ocr_file;

pub use engine::DetOnlyEngine;

pub use engine::OcrEngine;

pub use engine::OcrEngineConfig;

pub use engine::OcrResult_;

pub use engine::RecOnlyEngine;

pub use error::OcrError;

pub use error::OcrResult;

pub use mnn::Backend;

pub use mnn::InferenceConfig;

pub use mnn::InferenceEngine;

pub use mnn::PrecisionMode;

pub use postprocess::TextBox;

pub use rec::RecModel;

pub use rec::RecOptions;

pub use rec::RecognitionResult;

Modules

det

Text Detection Model

engine

OCR Engine

error

OCR error type definitions

mnn

MNN Inference Engine FFI Binding Layer

postprocess

Postprocessing Utilities

preprocess

Image Preprocessing Utilities

rec

Text Recognition Model

Functions

mnn_version

Get MNN version

version

Get library version