app.py

import streamlit as st
from transformers import AutoProcessor, AutoModelForVision2Seq
from pdf2image import convert_from_path
import base64
import io
from PIL import Image

# Load the OCR model and processor from Hugging Face
try:
    processor = AutoProcessor.from_pretrained("allenai/olmOCR-7B-0225-preview")
    model = AutoModelForVision2Seq.from_pretrained("allenai/olmOCR-7B-0225-preview")
except ImportError as e:
    processor = None
    model = None
    print(f"Error loading model: {str(e)}. Please ensure PyTorch is installed.")
except ValueError as e:
    processor = None
    model = None
    print(f"Error with model configuration: {str(e)}")

def process_pdf(pdf_file):
    """ Process the uploaded PDF file one page at a time, yielding HTML for each page with its image and extracted text. """
    if processor is None or model is None:
        return "<p>Error: Model could not be loaded. Check environment setup (PyTorch may be missing) or model compatibility.</p>"

    # Check if a PDF file was uploaded
    if pdf_file is None:
        return "<p>Please upload a PDF file.</p>"

    # Convert PDF to images
    try:
        pages = convert_from_path(pdf_file.name)
    except Exception as e:
        return f"<p>Error converting PDF to images: {str(e)}</p>"

    # Initial HTML with "Copy All" button and container for pages
    html = '<div><button onclick="copyAll()" style="margin-bottom: 10px;">Copy All</button></div><div id="pages">'

    # Process each page incrementally
    for i, page in enumerate(pages):
        # Convert the page image to base64 for embedding in HTML
        buffered = io.BytesIO()
        page.save(buffered, format="PNG")
        img_str = base64.b64encode(buffered.getvalue()).decode()
        img_data = f"data:image/png;base64,{img_str}"

        # Extract text from the page using the OCR model
        try:
            inputs = processor(text="Extract the text from this image.", images=page, return_tensors="pt")
            outputs = model.generate(**inputs)
            text = processor.decode(outputs[0], skip_special_tokens=True)
        except Exception as e:
            text = f"Error extracting text: {str(e)}"

        # Generate HTML for this page's section
        textarea_id = f"text{i+1}"
        page_html = f'''
        <div class="page" style="margin-bottom: 20px; border-bottom: 1px solid #ccc; padding-bottom: 20px;">
            <h3>Page {i+1}</h3>
            <div style="display: flex; align-items: flex-start;">
                <img src="{img_data}" alt="Page {i+1}" style="max-width: 300px; margin-right: 20px;">
                <div style="flex-grow: 1;">
                    <textarea id="{textarea_id}" rows="10" style="width: 100%;">{text}</textarea>
                    <button onclick="copyText('{textarea_id}')" style="margin-top: 5px;">Copy</button>
                </div>
            </div>
        </div>
        '''
        html += page_html

    # After all pages are processed, close the div and add JavaScript
    html += '</div>'
    html += '''
    <script>
    function copyText(id) {
        var text = document.getElementById(id);
        text.select();
        document.execCommand("copy");
    }
    function copyAll() {
        var texts = document.querySelectorAll("#pages textarea");
        var allText = Array.from(texts).map(t => t.value).join("\\n\\n");
        navigator.clipboard.writeText(allText);
    }
    </script>
    '''

    return html

# Define the Streamlit interface
st.title("PDF Text Extractor")
st.markdown("Upload a PDF file and click 'Extract Text' to see each page's image and extracted text incrementally.")

pdf_input = st.file_uploader("Upload PDF", type=["pdf"])
submit_btn = st.button("Extract Text")

if submit_btn and pdf_input:
    output_html = process_pdf(pdf_input)
    st.components.v1.html(output_html, height=800)