Extract Text (RapidOCR)

@rapidai1 VersionExtract

Runs fast ONNX-based RapidOCR (PP-OCRv5) detection and recognition over each Image, emitting either a joined String transcript or per-line regions with Polygon<Double> or Rectangle<Double> bounds and optional confidence.

How it fits

Typical backends

OCR scene images from files and deliver the transcript and geometry over HTTP.

Pull frames over HTTP, OCR them, and extract named entities from the recognised text.

Caveats

I/O contractThe output arm is fixed when the backend deploys: a text consumer pins the String transcript, a region consumer pins per-line records with Polygon<Double> or Rectangle<Double> bounds. The polygon is RapidOCR's native quad and the rectangle its axis-aligned bound.
Parameter interactionuse_gpu routes the ONNX Runtime engine onto the GPU; without a compatible accelerator the engine runs on CPU. The detection and recognition stages are pinned to a matched PP-OCRv5 server model set.
CompatibilityThe PP-OCRv5 models and their character dictionary are taken from model_repo at model_revision, cached when the backend is built and loaded locally; nothing is fetched over the network at run time.
LimitationThe PP-OCRv5 model family is pinned; model_repo and model_revision only re-point the cached source, and the execution device is the remaining runtime knob.
AccuracyRapidOCR favours speed; for the highest accuracy on dense documents prefer a heavier engine, and deskew rotated scans upstream.
State lifetimeThe engine is built once at startup; changing the device requires a redeploy.

Versions

25cb1519latestdefault
Automated release
6/25/2026