Crate rlx_ocr 

OCR engine for RLX — native compiled detection + recognition graphs.

Load HuggingFace robertknight/ocrs weights as .safetensors (use rlx-ocr-convert with feature convert-rten to export from legacy .rten).

Detection + recognition run on every standard RLX backend when the matching rlx-runtime feature is enabled (cpu, metal, mlx, cuda, rocm, gpu, vulkan). Build with all-backends for a single binary that accepts all --device values.

Parity / baseline: parity-ocrs (upstream ocrs + RTen) or rten-inference (RTen graphs only).

Re-exports

pub use capabilities::validate_device;

pub use config::DEFAULT_ALPHABET;

pub use config::DecodeMethod;

pub use config::DetectionParams;

pub use config::OcrConfig;

pub use ctc::CtcHypothesis;

pub use ctc::CtcStep;

pub use ctc::decode;

pub use engine::OcrEngine;

pub use engine::OcrEngineParams;

pub use engine::OcrInput;

pub use engine::input_image;

pub use engine::ocr_rgb_bytes;

pub use preprocess::BLACK_VALUE;

pub use preprocess::DimOrder;

pub use preprocess::ImageSource;

pub use preprocess::ImageSourceError;

pub use preprocess::prepare_image;

pub use runner::OcrOutput;

pub use runner::OcrRunner;

pub use runner::OcrRunnerBuilder;

pub use text::TextChar;

pub use text::TextItem;

pub use text::TextLine;

pub use text::TextWord;

pub use weights::HF_DETECTION_RTEN;

pub use weights::HF_DETECTION_ST;

pub use weights::HF_DETECTION_ST_FULL;

pub use weights::HF_RECOGNITION_RTEN;

pub use weights::HF_RECOGNITION_ST;

pub use weights::HF_RECOGNITION_ST_FULL;

pub use weights::SafetensorsFile;

pub use weights::is_rten_checkpoint;

pub use weights::load_rlx_weights;

pub use weights::load_safetensors;

pub use weights::load_safetensors_weights;

pub use weights::resolve_model_dir;

pub use model::DetectionGraphConfig;

pub use model::RecognitionGraphConfig;

pub use rlx::RlxTextDetector;

pub use rlx::RlxTextRecognizer;

Modules

capabilities

Backend capability checks for OCR graphs.

cli

config

OCR configuration — detection thresholds, alphabet, decode method.

ctc

CTC decoding (greedy + beam search).

detection

engine

High-level OCR engine — detect, layout, recognize.

geom

Geometry helpers for line polygons and edge extraction.

layout

Layout analysis — group detected words into reading-order lines (upstream ocrs algorithm).

model

Native RLX graphs for ocrs detection (U-Net) and recognition (CRNN + GRU).

preprocess

Greyscale image preprocessing for OCR models.

recognition

Recognition module — line batching helpers and full CRNN/GRU inference.

rlx

Compiled RLX inference for ocrs detection and recognition.

runner

High-level OCR runner with image loading.

text

Recognized text items with character-level bounding boxes.

weights

Weight paths, safetensors load, and optional .rten export.

Structs

RotatedRect

An oriented rectangle.

Constants

STANDARD_DEVICES

Backends every model crate is expected to support when the matching rlx-runtime feature is enabled at build time.

STANDARD_DEVICE_NAMES

CLI / help string for --device.