Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
from transformers import TrOCRProcessor, VisionEncoderDecoderModel, pipeline
|
| 3 |
+
from PIL import Image
|
| 4 |
+
|
| 5 |
+
# Load OCR model (TrOCR for handwritten documents)
|
| 6 |
+
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
|
| 7 |
+
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
|
| 8 |
+
|
| 9 |
+
# Optional: grammar correction model
|
| 10 |
+
corrector = pipeline("text2text-generation", model="vennify/t5-base-grammar-correction")
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
def digitize_image(image):
|
| 14 |
+
# Step 1: OCR
|
| 15 |
+
pixel_values = processor(images=image, return_tensors="pt").pixel_values
|
| 16 |
+
generated_ids = model.generate(pixel_values)
|
| 17 |
+
raw_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
| 18 |
+
|
| 19 |
+
# Step 2: Cleanup (optional)
|
| 20 |
+
cleaned_text = corrector(raw_text, max_length=512)[0]['generated_text']
|
| 21 |
+
|
| 22 |
+
return f"📝 Raw OCR:\n{raw_text}\n\n✅ Cleaned Text:\n{cleaned_text}"
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
# Gradio UI
|
| 26 |
+
demo = gr.Interface(
|
| 27 |
+
fn=digitize_image,
|
| 28 |
+
inputs=gr.Image(type="pil"),
|
| 29 |
+
outputs="text",
|
| 30 |
+
title="📜 Digitize Historical Records with AI",
|
| 31 |
+
description="Upload a scanned historical document (handwritten or printed). This app uses Hugging Face TrOCR to extract text and optionally clean it using a grammar-aware language model."
|
| 32 |
+
)
|
| 33 |
+
|
| 34 |
+
if __name__ == "__main__":
|
| 35 |
+
demo.launch()
|