app.py

import gradio as gr
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
import torch
import spaces

# Load TrOCR model
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-large-handwritten")
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-large-handwritten")

@spaces.GPU
def recognize_text(image):
    try:
        # Convert image to RGB if it's not already
        image = image.convert("RGB")
        print("Image converted to RGB.")

        # Preprocess the image
        pixel_values = processor(images=image, return_tensors="pt").pixel_values
        print("Image preprocessed. Pixel values shape:", pixel_values.shape)

        # Generate text from the image
        with torch.no_grad():  # Disable gradient calculation for inference
            generated_ids = model.generate(pixel_values)
            print("Generated IDs:", generated_ids)

        # Decode the generated IDs to text
        text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
        print("Decoded text:", text)

        return text
    except Exception as e:
        print(f"Error: {str(e)}")
        return f"Error: {str(e)}"

# Gradio UI
note = gr.Interface(
    fn=recognize_text,
    inputs=gr.Image(type="pil"),
    outputs="text",
    title="Handwritten Note to Digital Text",
    description="Upload an image of handwritten text, and the AI will convert it to digital text."
)

note.launch()