Skip to content

api-overview.md

Latest commit

 

History

History
103 lines (80 loc) · 2.97 KB

api-overview.md

File metadata and controls

103 lines (80 loc) · 2.97 KB
Raw

API Overview

← Back to README

Modules

  • lib.rs — OcrEngine (detection + recognition) + DocOrientationClassifier
  • det.rs — Text detection via DBNet
  • rec.rs — Text recognition via CRNN
  • decode.rs — CTC greedy decoding
  • cls.rs — Document orientation classification (PP-LCNet)

Main Structs

Struct Description Docs
OcrEngine OCR engine (detection + recognition) ocr-engine.md
DocOrientationClassifier Document orientation classifier doc-orientation.md

Return Types

TextRegion

pub struct TextRegion {
    pub bbox: [[f32; 2]; 4],   // Four corner coordinates [[x1,y1], [x2,y2], [x3,y3], [x4,y4]]
    pub confidence: f32,        // Confidence (0.0 ~ 1.0)
}

OcrBlock

pub struct OcrBlock {
    pub text: String,           // Recognized text
    pub confidence: f32,        // Confidence (0.0 ~ 1.0)
    pub x: f32,                 // Bounding box top-left x coordinate
    pub y: f32,                 // Bounding box top-left y coordinate
    pub width: f32,             // Bounding box width
    pub height: f32,            // Bounding box height
}

DecodedText

pub struct DecodedText {
    pub text: String,           // Decoded text
    pub score: f32,             // Average confidence
}

OrientationResult

pub struct OrientationResult {
    pub orientation: DocOrientation,  // Detected orientation
    pub confidence: f32,              // Confidence (0.0 ~ 1.0)
}

DocOrientation

pub enum DocOrientation {
    Upright,    // 0° - Normal orientation
    Rotate90,   // 90° - Needs clockwise rotation of 90°
    Rotate180,  // 180° - Needs rotation of 180°
    Rotate270,  // 270° - Needs counterclockwise rotation of 90°
}

OrderBy

pub enum OrderBy {
    Horizontal,  // Arrange horizontally (top to bottom, left to right)
    Vertical,    // Arrange vertically (right to left, top to bottom)
    Score,       // Arrange by confidence descending
}

AccelerationDevice

Hardware acceleration device for ONNX Runtime inference.

pub enum AccelerationDevice {
    Cpu,        // CPU-only inference (default, always available)
    DirectML,   // DirectML acceleration (Windows, DirectX 12 GPU)
    Cuda,       // CUDA acceleration (NVIDIA GPU)
    OpenVINO,   // Intel OpenVINO acceleration (Windows/Linux)
    Nnapi,      // Android NNAPI acceleration (Android)
    Coreml,     // Apple CoreML acceleration (macOS/iOS)
    Cann,       // Huawei CANN / Ascend NPU acceleration (Linux)
}

Key methods:

  • is_available(&self) -> bool — Check if the device is available (compile-time feature + runtime EP check)
  • available_devices() -> Vec<AccelerationDevice> — List all available devices in the current environment

See acceleration.md for detailed platform requirements, availability checking, and runtime behavior.