README.md

---
license: apache-2.0
tags:
- deepfake-detection
- multimodal
- image-classification
- text-classification
- ensemble
- gradcam
- explainability
datasets:
- Hemg/deepfake-and-real-images
- artem9k/ai-text-detection-pile
pipeline_tag: image-classification
---

# 🕵️ Multimodal Deepfake Detection Model

Modèle d'ensemble multimodal pour détecter les contenus générés par IA (images, vidéos, texte) avec explicabilité GradCAM.

## 🏗️ Architecture

```
Visual Branch    EfficientNet-B0/B4  ──┐
                                       ├──► Fusion pondérée ──► Confidence [0-1]
Text Branch      RoBERTa-base         ──┘
                    ↓
               GradCAM Heatmap (explicabilité)
```

## 🚀 Utilisation Rapide

```python
from inference import load_model, classify_image, classify_text, classify_multimodal

model, config = load_model('multimodal_ensemble.pt', device='cuda')

# Image + GradCAM
result = classify_image(model, 'face.jpg', return_gradcam=True)
print(f"{result['prediction']} — confidence: {result['confidence']:.2%}")
# result['gradcam'] → heatmap (224, 224)

# Texte
result = classify_text(model, 'Cet essai analyse les impacts...')
print(f"{result['prediction']} — confidence: {result['confidence']:.2%}")

# Multimodal (image + texte)
result = classify_multimodal(model, image_path_or_pil='face.jpg', text='Caption...')
print(f"Fusion: {result['prediction']} — poids: {result['fusion_weights']}")
```

## 📊 Jeux de Données
- **Visuel**: [Hemg/deepfake-and-real-images](https://huggingface.co/datasets/Hemg/deepfake-and-real-images) — 528K images
- **Texte**: [artem9k/ai-text-detection-pile](https://huggingface.co/datasets/artem9k/ai-text-detection-pile) — 1.88GB

## 📁 Fichiers
| Fichier | Description |
|---------|-------------|
| `model.py` | Architecture complète |
| `preprocessing.py` | Pipeline de données |
| `inference.py` | API d'inférence |
| `train.py` / `train_optimised.py` | Scripts d'entraînement |
| `multimodal_ensemble.pt` | Checkpoint principal |
| `gradcam_examples/` | Visualisations d'explicabilité |

## 🔗 Space de Démo
[alianassmaaa/multimodal-deepfake-space](https://huggingface.co/spaces/alianassmaaa/multimodal-deepfake-space)

## 📄 Licence
Apache-2.0