app.py

from fastapi import FastAPI, HTTPException, File, UploadFile
from fastapi.middleware.cors import CORSMiddleware
import torch
import torchvision.transforms as transforms
from PIL import Image
from pytorch_grad_cam import GradCAM
from pytorch_grad_cam.utils.image import show_cam_on_image
from transformers import AutoFeatureExtractor
import timm
import numpy as np
import json
import base64
from io import BytesIO
import uvicorn

app = FastAPI(title="VerifAI GradCAM API", description="API pour la détection d'images IA avec GradCAM")

# Configuration CORS
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

class AIDetectionGradCAM:
    def __init__(self):
        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
        self.models = {}
        self.feature_extractors = {}
        self.target_layers = {}
        
        # Initialiser les modèles
        self._load_models()
    
    def _load_models(self):
        """Charge les modèles pour la détection"""
        try:
            # Modèle Swin Transformer
            model_name = "microsoft/swin-base-patch4-window7-224-in22k"
            self.models['swin'] = timm.create_model('swin_base_patch4_window7_224', pretrained=True, num_classes=2)
            self.feature_extractors['swin'] = AutoFeatureExtractor.from_pretrained(model_name)
            
            # Définir les couches cibles pour GradCAM
            self.target_layers['swin'] = [self.models['swin'].layers[-1].blocks[-1].norm1]
            
            # Mettre en mode évaluation
            for model in self.models.values():
                model.eval()
                model.to(self.device)
                
        except Exception as e:
            print(f"Erreur lors du chargement des modèles: {e}")
    
    def _preprocess_image(self, image, model_type='swin'):
        """Prétraite l'image pour le modèle"""
        if isinstance(image, str):
            # Si c'est un chemin ou base64
            if image.startswith('data:image'):
                # Décoder base64
                header, data = image.split(',', 1)
                image_data = base64.b64decode(data)
                image = Image.open(BytesIO(image_data))
            else:
                image = Image.open(image)
        
        # Convertir en RGB si nécessaire
        if image.mode != 'RGB':
            image = image.convert('RGB')
        
        # Redimensionner
        image = image.resize((224, 224))
        
        # Normalisation standard
        transform = transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], 
                               std=[0.229, 0.224, 0.225])
        ])
        
        tensor = transform(image).unsqueeze(0).to(self.device)
        return tensor, np.array(image) / 255.0
    
    def _generate_gradcam(self, image_tensor, rgb_img, model_type='swin'):
        """Génère la carte de saillance GradCAM"""
        try:
            model = self.models[model_type]
            target_layers = self.target_layers[model_type]
            
            # Créer l'objet GradCAM
            cam = GradCAM(model=model, target_layers=target_layers)
            
            # Générer la carte de saillance
            grayscale_cam = cam(input_tensor=image_tensor, targets=None)
            grayscale_cam = grayscale_cam[0, :]
            
            # Superposer sur l'image originale
            cam_image = show_cam_on_image(rgb_img, grayscale_cam, use_rgb=True)
            
            return cam_image
            
        except Exception as e:
            print(f"Erreur GradCAM: {e}")
            return rgb_img * 255
    
    def predict_and_explain(self, image):
        """Prédiction avec explication GradCAM"""
        try:
            # Prétraitement
            image_tensor, rgb_img = self._preprocess_image(image)
            
            # Prédiction
            with torch.no_grad():
                outputs = self.models['swin'](image_tensor)
                probabilities = torch.nn.functional.softmax(outputs, dim=1)
                confidence = probabilities.max().item()
                prediction = probabilities.argmax().item()
            
            # Génération GradCAM
            cam_image = self._generate_gradcam(image_tensor, rgb_img)
            
            # Convertir l'image GradCAM en base64
            pil_image = Image.fromarray(cam_image.astype(np.uint8))
            buffer = BytesIO()
            pil_image.save(buffer, format='PNG')
            cam_base64 = base64.b64encode(buffer.getvalue()).decode()
            
            # Résultats
            result = {
                'prediction': prediction,
                'confidence': confidence,
                'class_probabilities': {
                    'Real': probabilities[0][0].item(),
                    'AI-Generated': probabilities[0][1].item()
                },
                'cam_image': f"data:image/png;base64,{cam_base64}",
                'status': 'success'
            }
            
            return result
            
        except Exception as e:
            return {'status': 'error', 'message': str(e)}

# Initialiser le détecteur
detector = AIDetectionGradCAM()

@app.get("/")
async def root():
    return {"message": "VerifAI GradCAM API", "status": "running"}

@app.post("/predict")
async def predict_image(file: UploadFile = File(...)):
    """Endpoint pour analyser une image"""
    try:
        # Lire l'image
        image_data = await file.read()
        image = Image.open(BytesIO(image_data))
        
        # Analyser
        result = detector.predict_and_explain(image)
        
        return result
        
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/predict-base64")
async def predict_base64(data: dict):
    """Endpoint pour analyser une image en base64"""
    try:
        if 'image' not in data:
            raise HTTPException(status_code=400, detail="Champ 'image' requis")
        
        image_b64 = data['image']
        
        # Analyser
        result = detector.predict_and_explain(image_b64)
        
        return result
        
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)