app.py

import gradio as gr
import torch
from transformers import LayoutLMv3Processor, LayoutLMv3ForTokenClassification
import pytesseract

# Set the Tesseract executable path (for Windows users)
import pytesseract
pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"


# Load the model and processor
processor = LayoutLMv3Processor.from_pretrained("quadranttechnologies/Table_OCR")
model = LayoutLMv3ForTokenClassification.from_pretrained("quadranttechnologies/Table_OCR")
model.eval()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

def process_image(image):
    try:
        # Preprocess the image using the processor
        encoding = processor(image, return_tensors="pt", truncation=True, padding="max_length", max_length=512)

        # Move inputs to the same device as the model
        encoding = {key: val.to(device) for key, val in encoding.items()}

        # Perform inference
        with torch.no_grad():
            outputs = model(**encoding)
            predictions = torch.argmax(outputs.logits, dim=-1)

        # Extract input IDs, bounding boxes, and predicted labels
        words = encoding["input_ids"]
        bboxes = encoding["bbox"]
        labels = predictions.squeeze().tolist()

        # Format output as JSON
        structured_output = []
        for word_id, bbox, label in zip(words.squeeze().tolist(), bboxes.squeeze().tolist(), labels):
            # Decode the word ID to text
            word = processor.tokenizer.decode([word_id]).strip()
            if word:  # Avoid adding empty words
                structured_output.append({
                    "word": word,
                    "bounding_box": bbox,
                    "label": model.config.id2label[label]  # Convert label ID to label name
                })

        return structured_output

    except Exception as e:
        return {"error": str(e)}  # Return error details if any issue occurs

# Define the Gradio interface
interface = gr.Interface(
    fn=process_image,
    inputs=gr.Image(type="pil"),  # Accepts image input
    outputs="json",  # Outputs JSON structure
    title="Table OCR",
    description="Upload an image (e.g., receipt or document) to extract structured information in JSON format."
)

# Launch the app
if __name__ == "__main__":
    interface.launch(share=True)