rec_chinese_char

A lightweight image-based Chinese character recognition model.

This model is fine-tuned from PaddleOCR/ch_ppocr_server_v2.0_rec_pre to focus on single-character classification, useful for applications like character-level OCR, glyph recognition.

Model Details

• Base: ch_ppocr_server_v2.0_rec_pre
• Task: Single Chinese character recognition
• Framework: PaddleOCR / PaddlePaddle
• Output: Unicode character predicted from input image

Files

File	Description
inference.pdmodel	Paddle model structure
inference.pdiparams	Model weights
inference.pdiparams.info	Parameter metadata (optional, for Paddle)
inference.yml	Model config file (used by PaddleOCR tools)
rec_custom_keys.txt	Character dictionary used during training
README.md	This documentation file

Usage

You can use this model with PaddleOCR:

import paddle
from paddleocr import PaddleOCR

gpu_available = paddle.device.is_compiled_with_cuda()
ocr = PaddleOCR(
    use_angle_cls=False,
    lang='ch',
    det=False,
    use_gpu=gpu_available,
    rec_model_dir='your_model_dir'  # path to this model
    rec_char_dict_path='your_model_dir/rec_custom_keys.txt',
    rec_image_shape='3,48,48',
)

result = ocr.ocr('path/to/image.jpg')

Make sure to use the correct label dictionary (rec_custom_keys.txt).

Notes

This model is intended for small-scale, character-level recognition tasks.