PPOCR

PPOCR is a PyTorch implementation of the PP-OCR model.

PP-OCR models, originally developed by Baidu as part of the PaddleOCR library and implemented in PaddlePaddle, are known for their speed and lightweight design. In my experience, they have outperformed other OCR solutions such as Tesseract.

This library provides a PyTorch-based version of the original PP-OCR model.

Currently, only the PP-OCRv3 model is supported.

Installation

pip install ppocr

Usage

from ppocr.ocr import PPOCR
from PIL import Image
import numpy as np

ocr = PPOCR()

image = Image.open("path/to/image.jpg")
image = np.array(image)

results = ocr.predict(image)

for result in results:
    print(result)

# Output:
# [
#     OCRResult(
#         text=...,
#         confidence=...,
#         box=Box2D(top_left=Point2D(x, y), bottom_right=Point2D(x, y)),
#     ),
#     OCRResult(..),
#     ...
# ]

Credits

• PaddleOCR
• PaddleOCR2Pytorch

TODO

• [] Add support for PP-OCRv4