app.py

import gradio as gr
from transformers import DonutProcessor, VisionEncoderDecoderModel
from PIL import Image
import torch
import json

# ===============================
# Charger le modèle Donut public
# ===============================
model_name = "naver-clova-ocr-donut-base"
processor = DonutProcessor.from_pretrained(model_name, revision="main")
model = VisionEncoderDecoderModel.from_pretrained(model_name, revision="main")

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

# ===============================
# Fonction extraction colonne Description
# ===============================
def extract_description(image_pil):
    # Préparer l'image
    pixel_values = processor(images=image_pil, return_tensors="pt").pixel_values.to(device)

    # Générer le texte
    generated_ids = model.generate(pixel_values, max_length=1024)
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

    # Donut renvoie souvent du JSON ou semi-structuré
    try:
        data = json.loads(generated_text)
    except:
        data = {"text": generated_text}

    # Extraire les lignes contenant "Description"
    desc_lines = []
    if isinstance(data, dict):
        for key, value in data.items():
            if "description" in key.lower():
                if isinstance(value, list):
                    desc_lines.extend(value)
                else:
                    desc_lines.append(str(value))
    else:
        # fallback si Donut ne renvoie pas JSON
        lines = generated_text.split("\n")
        found_header = False
        for line in lines:
            if found_header:
                desc_lines.append(line)
            elif "description" in line.lower():
                found_header = True

    if not desc_lines:
        return "❌ Colonne 'Description' non trouvée", generated_text
    else:
        return "\n".join(desc_lines), generated_text

# ===============================
# Interface Gradio
# ===============================
demo = gr.Interface(
    fn=extract_description,
    inputs=gr.Image(type="pil", label="Image de facture"),
    outputs=[
        gr.Textbox(label="📋 Colonne Description"),
        gr.Textbox(label="🛠 Texte complet Donut")
    ],
    title="Extraction de la colonne Description (Donut)",
    description="Détection automatique de la colonne Description dans les factures avec Donut"
)

demo.launch()