Crate df_ocr_switcher 

df-ocr-switcher — pipeline OCR documento in puro Rust.

Architettura

Input (TIFF/PNG/JPEG)
  └── DocPipeline
        ├── DocOrientationClassifier (ppocr-rs PP-LCNet)  → rotazione
        ├── LayoutAnalyzer           (ppocr-rs PP-DocLayoutV3) → regioni
        └── OcrEngine (impl):
              ├── PpOcrEngine   → detect_with_layout (PP-OCRv6)
              └── TesseractEngine → Ocr5Engine + layout assignment
  └── Output
        ├── Markdown  (to_markdown)
        └── JSON      (serde_json)

Quickstart

use df_ocr_switcher::{DocPipeline, OutputFormat, PpOcrEngineConfig, engine::ppocr::PpOcrEngine};
use std::path::PathBuf;

let engine = PpOcrEngine::new(PpOcrEngineConfig::default()).unwrap();
let mut pipeline = DocPipeline::new(
    Box::new(engine),
    "models/paddleocr/layout/PP-DocLayoutV3.onnx",
).unwrap();

let md = pipeline.process_file("scan.tiff", OutputFormat::Markdown).unwrap();
println!("{md}");

Re-exports

pub use engine::OcrBlock;

pub use engine::OcrEngine;

pub use engine::OcrFormula;

pub use engine::OcrLayoutBox;

pub use engine::OcrPageResult;

pub use engine::OcrTable;

pub use engine::OcrWord;

pub use engine::ppocr::OcrModelPaths;

pub use engine::ppocr::PpOcrEngine;

pub use engine::ppocr::PpOcrEngineConfig;

pub use engine::ppocr::TableModelPaths;

pub use error::Error;

pub use error::Result;

pub use output::markdown::page_to_markdown;

pub use output::markdown::to_markdown;

pub use pipeline::DocPipeline;

pub use pipeline::OutputFormat;

Modules

engine

Trait comune e tipi condivisi per tutti gli engine OCR.

error

input

output

pipeline

DocPipeline — orchestratore principale.