src/streamlit_app.py

import streamlit as st
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
import torch

# Configuration de la page
st.set_page_config(page_title="OCR Manuscrit avec TrOCR", layout="centered")

st.title("✍️ OCR de texte manuscrit avec TrOCR")
st.write("Chargez une image contenant du texte manuscrit pour en extraire le contenu.")

# Chargement du modèle TrOCR pour manuscrit
@st.cache_resource
def load_model():
    processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
    model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
    return processor, model

processor, model = load_model()

# Upload de l'image
uploaded_file = st.file_uploader("📤 Charger une image manuscrite (format .png ou .jpg)", type=["png", "jpg", "jpeg"])

if uploaded_file is not None:
    image = Image.open(uploaded_file).convert("RGB")
    st.image(image, caption="🖼️ Image chargée", use_column_width=True)

    if st.button("🔍 Lancer la reconnaissance"):
        with st.spinner("Reconnaissance en cours..."):
            pixel_values = processor(images=image, return_tensors="pt").pixel_values
            generated_ids = model.generate(pixel_values)
            generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

        st.success("✅ Texte reconnu :")
        st.text_area("📝 Résultat OCR", generated_text, height=150)