app.py

import os
os.environ['USE_TORCH'] = '1'

from doctr.io import DocumentFile
from doctr.models import ocr_predictor
import gradio as gr
from PIL import Image
import base64
from utils import HocrParser
import google.generativeai as genai

api_key = 'AIzaSyB4lQJM89q1EQDNtKoh8E5wf-ks0E6Q_Uc' # put your API key here
genai.configure(api_key=api_key)
geminiModel = genai.GenerativeModel(model_name='gemini-2.0-flash')
predictor = ocr_predictor(det_arch='db_mobilenet_v3_large', reco_arch='crnn_vgg16_bn',pretrained=True)
predictor.reco_predictor.model.cfg['vocab']='0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ!”#$%&’()*+,-./:;<=>?@[]^_`{|}~°£€¥¢฿áàảạãăắằẳẵặâấầẩẫậéèẻẽẹêếềểễệóòỏõọôốồổộỗơớờởợỡúùủũụưứừửữựiíìỉĩịýỳỷỹỵÁÀẢẠÃĂẮẰẲẴẶÂẤẦẨẪẬÉÈẺẼẸÊẾỀỂỄỆÓÒỎÕỌÔỐỒỔỘỖƠỚỜỞỢỠÚÙỦŨỤƯỨỪỬỮỰIÍÌỈĨỊÝỲỶỸỴ'
title="DocTR OCR (PDL Demo)"
description="Upload an image to get the OCR results !"

def greet(img):
    img.save("out.jpg")
    doc = DocumentFile.from_images("out.jpg")
    output=predictor(doc)

    xml_outputs = output.export_as_xml()
    parser = HocrParser()

    res=""
    for obj in output.pages:
      for obj1 in obj.blocks:
        for obj2 in obj1.lines:
          for obj3 in obj2.words:
            res=res + " " + obj3.value
        res=res + "\n"
      res=res + "\n"
    # generate text
    prompt = "the input text in vietnamese, please add accend and take this peace of information and give all the information in point wise better format also give some recomendation related to them: " + res
    # print(prompt)
    response  = geminiModel.generate_content(prompt)
    print(response)
    res = response.text
    _output_name = "RESULT_OCR.txt"
    _output_name_pdf="RESULT_OCR.pdf"

    open(_output_name, 'w').close() # clear file
    with open(_output_name, "w", encoding="utf-8", errors="ignore") as f:
        f.write(res)
        print("Writing into file")

    base64_encoded_pdfs = list()
    for i, (xml, img) in enumerate(zip(xml_outputs, doc)):
      xml_element_tree = xml[1]
      parser.export_pdfa(_output_name_pdf,
            hocr=xml_element_tree, image=img)
      with open(_output_name_pdf, 'rb') as f:
            base64_encoded_pdfs.append(base64.b64encode(f.read()))
    return res, _output_name, _output_name_pdf

demo = gr.Interface(fn=greet,
                    inputs=gr.Image(type="pil"),
                    outputs=["text", "file","file"],
                    title=title,
                    description=description
                    )

demo.launch(debug=True)