The Face Attractiveness Detection Model leverages deep learning to assess facial attractiveness. Built with Streamlit, the app allows users to upload images, videos, or use live cameras for real-time evaluation. The model uses pretrained neural network weights from Google Drive, downloaded via gdown. The FaceAttractivenessApp module handles core processing, while Streamlit ensures an intuitive interface. The architecture supports diverse input formats and accurate predictions.
- Dataset: CelebA Dataset on Kaggle
- Kaggle Notebook: Face Attraction Detection Model
| Category | Technologies |
|---|---|
| Languages | Python |
| Libraries/Frameworks | TensorFlow, Keras, OpenCV, MTCNN, PIL, NumPy, gdown |
| Cloud Storage | Google Drive |
| Tools | GitHub, Git, VS Code |
| Development Environment | Jupyter Notebook |
| Deployment | Streamlit |
The model predicts facial attractiveness from static images, video feeds, or live camera streams via a deep learning pipeline.
The dataset includes 40 features and 202,599 entries.
| Feature | Description | Values |
|---|---|---|
| 5_o_Clock_Shadow | Shadow near chin and jawline | 0 (No), 1 (Yes) |
| Arched_Eyebrows | Curved eyebrows | 0 (No), 1 (Yes) |
| Attractive | Aesthetic appeal | 0 (No), 1 (Yes) |
| Bags_Under_Eyes | Dark circles or puffiness | 0 (No), 1 (Yes) |
| Bald | No hair on scalp | 0 (No), 1 (Yes) |
| Smiling | Smile expression | 0 (No), 1 (Yes) |
| Young | Youthful appearance | 0 (No), 1 (Yes) |
(Additional features omitted for brevity)
- Data Preprocessing: Clean and augment the dataset.
- Model Training: Train Inception V3 deep learning model.
- Evaluation: Assess performance and accuracy.
- Model Generation: Save as a
.kerasfile. - Conversion: Convert to TensorFlow Lite format.
- Deployment: Integrate with Streamlit for user interaction.
Inception V3 shows promising performance with 86.80% accuracy.
