app.py

import torch
import cv2
import numpy as np
import gradio as gr
from PIL import Image
from transformers import TrOCRProcessor, VisionEncoderDecoderModel

# 1. Setup 
device = "cpu"
processor = TrOCRProcessor.from_pretrained('microsoft/trocr-large-handwritten')
model = VisionEncoderDecoderModel.from_pretrained('microsoft/trocr-large-handwritten').to(device)

def get_lines_from_image(img_array):
    # Gradio gives us a numpy array directly
    gray = cv2.cvtColor(img_array, cv2.COLOR_RGB2GRAY)
    
    # Binary threshold
    _, thresh = cv2.threshold(gray, 180, 255, cv2.THRESH_BINARY_INV)

    # Dilate horizontally to join letters into lines
    kernel = np.ones((5, 100), np.uint8) 
    dilation = cv2.dilate(thresh, kernel, iterations=1)

    # Find the boundaries of each line
    contours, _ = cv2.findContours(dilation, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    
    # Sort contours from top to bottom
    contours = sorted(contours, key=lambda ctr: cv2.boundingRect(ctr)[1])

    line_images = []
    for ctr in contours:
        x, y, w, h = cv2.boundingRect(ctr)
        if h > 15 and w > 15:
            # Add padding
            y_start, y_end = max(0, y-10), min(img_array.shape[0], y+h+10)
            x_start, x_end = max(0, x-10), min(img_array.shape[1], x+w+10)
            
            roi = img_array[y_start:y_end, x_start:x_end]
            line_images.append(Image.fromarray(roi)) # ROI is already RGB from Gradio
            
    return line_images

def full_pipeline(input_img):
    if input_img is None:
        return "Please upload an image."
        
    lines = get_lines_from_image(input_img)
    if not lines:
        return "No text lines detected. Try a clearer image."

    final_transcript = []
    for line_img in lines:
        pixel_values = processor(images=line_img, return_tensors="pt").pixel_values.to(device)
        
        with torch.no_grad():
            generated_ids = model.generate(pixel_values)
        
        text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
        
        # Cleanup
        text = text.strip().rstrip('. ').strip()
        final_transcript.append(text)

    return " ".join(final_transcript)

# Gradio UI Setup
demo = gr.Interface(
    fn=full_pipeline, 
    inputs=gr.Image(), 
    outputs="text",
    title="Handwritten Paragraph to Typed Text",
    description="Upload a handwritten paragraph."
)

if __name__ == "__main__":
    demo.launch()