Crate sceptre 

sceptre

A Rust reimplementation of EasyOCR’s OCR pipeline — CRAFT text detection followed by gen2 CRNN recognition with CTC decoding — running the models over ONNX.

Architecture

The pipeline is three stages behind a single Reader handle:

1. Detection ([detect]) — CRAFT produces region/link heat-maps that are post-processed into text boxes.
2. Recognition ([recognize]) — each box is cropped, normalized, and run through a gen2 CRNN; the logits are CTC-decoded into text.
3. Inference backend ([inference]) — a runtime-neutral [inference::ModelBackend] seam abstracts over ort (native) and tract (pure-Rust), with candle reserved as a deferred pure-Rust fallback.

Models are resolved and cached by [models]; behavior is configured through [config].

Structs

BBox

An axis-aligned bounding box [x_min, y_min, x_max, y_max].

ConcurrencyConfig

Caps the worker and inference thread pools within each reader.

DetectionConfig

Parameters controlling CRAFT detection and box grouping.

FallbackEngine

Composes engines, trying each in order until one returns a non-empty result.

Image

A decoded, owned RGB8 image — the public input DTO for the OCR engine.

ModelConfig

Model selection and provisioning configuration.

ModelDescriptor

Filesystem-free registry metadata for one model artifact.

ModelInfo

A model required by a configuration, with its resolved repo id and cache status.

OcrConfig

Top-level OCR configuration.

OcrResult

The full result of an OCR run over one image.

Point

A 2D point in pixel coordinates.

Quad

A four-point quadrilateral (clockwise from top-left), used for rotated text.

ReadOptions

Per-call options for a Reader::readtext or Reader::recognize invocation.

Reader

A ready-to-use OCR reader.

ReaderBuilder

Builder for Reader, filling injectable seams with in-crate defaults.

RecognitionConfig

Parameters controlling recognition and decoding.

TextLine

A single recognized line of text with its location and confidence.

VerifiedModelProvider

An in-memory provider that verifies every artifact before exposing it.

Enums

Backend

Which inference backend to use.

Decoder

CTC decoding strategy.

Language

A supported recognition language group (gen2 models only).

ModelArtifact

A model artifact resolved either as a local path or in-memory bytes.

ModelRole

The role a model plays in the pipeline.

OcrError

The single error type for all sceptre operations.

Constants

VERSION

The sceptre crate version.

Traits

ModelProvider

Resolves detector and recognizer model artifacts.

OcrEngine

The primary extension seam: an OCR engine turns an image into recognized text.

ProgressSink

Receives coarse progress notifications from the pipeline.

Functions

download_models

Ensure every model in config’s manifest is present locally, downloading any that are missing, and return the resulting (now-cached) manifest. Requires the download feature; without it the underlying fetch returns an crate::OcrError::Model.

model_descriptors

Describe the detector and each distinct configured recognizer without filesystem or network access.

model_manifest

The models required by config: the CRAFT detector plus one recognizer per configured language (in order, duplicates preserved), each annotated with cache status. Pure filesystem inspection — no network.

Type Aliases

Result

Standard result type for all fallible operations in sceptre.