Parameters
Generate a configuration file¶
This creates default_rapidocr.yaml in the current directory. You can pass the file with RapidOCR(config_path=...), or pass individual values through the params mapping.
Info
This page summarizes the most frequently used parameters. See the complete Chinese parameter reference for every field, default value and version-specific note.
Common parameter groups¶
Global¶
| Parameter | Description |
|---|---|
text_score |
Minimum text confidence used for output filtering. |
use_det, use_cls, use_rec |
Enable or disable detection, orientation classification and recognition. |
min_side_len, max_side_len |
Image resize limits used during preprocessing. |
use_vertical_padding |
Add vertical padding for narrow text regions. |
return_word_box |
Return word-level boxes when supported. |
model_root_dir |
Root directory for downloaded or local models. |
Det¶
| Parameter | Description |
|---|---|
engine_type |
Inference backend, such as onnxruntime, openvino, paddle, torch, mnn or tensorrt. |
lang_type |
Detection language route. The available values depend on the OCR version and model route. |
model_type |
Model size or variant, such as tiny, small, medium, mobile or server. |
ocr_version |
OCR model family, such as PP-OCRv4, PP-OCRv5 or PP-OCRv6. |
limit_side_len |
Target side length used by detection preprocessing. |
box_thresh |
Detection box threshold. A higher value usually reduces recall. |
unclip_ratio |
Expansion ratio applied to detected text boxes. |
Cls¶
| Parameter | Description |
|---|---|
cls_image_shape |
Input shape for orientation classification, default [3, 48, 192]. |
cls_batch_num |
Classification batch size. The default is usually sufficient. |
cls_thresh |
Confidence threshold for orientation correction. |
label_list |
Orientation labels, normally ["0", "180"]. Do not change unless using a compatible custom model. |
Rec¶
| Parameter | Description |
|---|---|
lang_type |
Recognition language route. See the model list. |
rec_img_shape |
Recognition input shape, default [3, 48, 320]. |
rec_batch_num |
Recognition batch size. |
rec_keys_path |
Optional dictionary file for a custom recognition model. |
For the complete parameter list, defaults and version-specific notes, see the Chinese parameter reference.