Usage

Overview¶

This guide covers the two main ways to use RapidOCR: the Python API and the rapidocr command-line interface. For installation, see the installation guide.

Starting with rapidocr>=3.10.0, model loading is lazy: detection, classification and recognition models are loaded when first used. The first call may therefore take longer than subsequent calls.

Info

This English page covers the primary workflow. The Chinese usage reference contains the complete stage-by-stage examples and return-value details.

Python usage¶

from rapidocr import RapidOCR

engine = RapidOCR()
image_url = "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg"
result = engine(image_url)

print(result.txts)
print(result.scores)
result.vis("vis_result.jpg")

You can pass a YAML configuration file:

rapidocr config

from rapidocr import RapidOCR

engine = RapidOCR(config_path="default_rapidocr.yaml")
result = engine(image_url)

You can also override individual settings through params:

from rapidocr import EngineType, ModelType, RapidOCR

engine = RapidOCR(
    params={
        "Det.engine_type": EngineType.OPENVINO,
        "Rec.model_type": ModelType.TINY,
    }
)

Image inputs¶

RapidOCR accepts a local path, URL, encoded bytes, NumPy array, Pillow image and supported memory buffers. See the image input reference for examples and buffer layout requirements.

Output¶

The return value is one of TextDetOutput, TextClsOutput, TextRecOutput or RapidOCROutput, depending on the enabled stages.

For the normal detection + classification + recognition pipeline, the most useful fields are:

Field	Description
`boxes`	Four-point coordinates for each detected text line.
`txts`	Recognized text, in the same order as `boxes`.
`scores`	Confidence score for each recognized line.
`word_results`	Word-level results when `return_word_box=True`.
`elapse_list`	Detection, classification and recognition times in seconds.
`elapse`	Total inference time in seconds.

print(result.boxes.shape)
print(result.txts)
print(result.scores)
print(result.elapse)

Set use_det, use_cls or use_rec to False when you need only selected stages. The corresponding return type and fields are documented in the Chinese usage reference.

CLI usage¶

Check the installation:

rapidocr check

Run OCR on an image and save a visualization:

rapidocr -img "https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/master/resources/test_files/ch_en_num.jpg" --vis_res

Download models for a configuration:

rapidocr download_models

For all available CLI options, run rapidocr --help.