RapidAI / RapidOCRPDF

Based on RapidOCR, extract the PDF content.
Apache License 2.0
118 stars 13 forks source link

调用GPU的rapid无法识别图片内容,但是调用CPU可以 #15

Open DeathGanker opened 4 months ago

DeathGanker commented 4 months ago
# image_ocr_processor.py
import base64
import os

from pathlib import Path
from utils import response
# 本地使用cpu资源进行识别
from rapidocr_onnxruntime import RapidOCR
# 生产环境使用GPU资源进行识别
# from rapidocr_paddle import RapidOCR

class ImageOCRProcessor:
    def __init__(self):
        self.engine = RapidOCR()  # 初始化OCR系统
        # 生产环境使用GPU资源进行识别
        # self.engine = RapidOCR(det_use_cuda=True, cls_use_cuda=True, rec_use_cuda=True)

    def process_image(self, image_input, use_det=True, use_cls=True, use_rec=True):
        # 根据输入类型处理图片
        if isinstance(image_input, (bytes, bytearray)) or isinstance(image_input, Path):
            # 如果输入是字节类型,则直接使用 engine 方法
            result, _ = self.engine(image_input, use_det=use_det, use_cls=use_cls, use_rec=use_rec)
        else:
            raise ValueError("Unsupported rapidocr_image input type")

        final_result = ""

        for item in result:
            text = item[1]
            final_result += " "+text

        # 将合并后的内容写入到文本文件中
        with open('output.txt', 'w', encoding='utf-8') as file:
            file.write(final_result)
        # 将output.txt文件转成文件流
        with open('output.txt', 'rb') as file:
            file_content = file.read()
            encoded_content = base64.b64encode(file_content).decode('utf-8')

        # os.remove('output.txt')

        response_result = response.success(encoded_content)
        json_data = response_result.to_dict()
        return json_data

# 为了在 Flask 应用中使用这个处理器,我们可以在这里直接创建一个实例并调用 process_image 方法
# 这里只是一个示例,您可以根据需要调整
if __name__ == '__main__':
    image_path = Path('../tests/test_files/2.jpg')
    image_processor = ImageOCRProcessor()
    ocr_result = image_processor.process_image(image_path)
    print(ocr_result)

主要区别在于import的init方法,在生产环境上使用GPU的调用方式,图片无法识别,转成CPU即可,可能是底层库的原因

DeathGanker commented 4 months ago

rapidocr-onnxruntime==1.3.15

paddleocr~=2.7.0.3

使用的版本

SWHL commented 3 months ago

我在百度AI Studio中测试rapidocr_paddle库,GPU端,没有问题。 建议先排查一下paddlepaddle库在生产环境下调用GPU有无问题,可以用以下代码来自查一下:

import paddle

print(paddle.utils.run_check())
# 如果出现PaddlePaddle is installed successfully!,说明您已成功安装。