Face Shape Classifier
Project Overview
AI-powered app that detects face shape from photo and suggests best hairstyle.
Model Details
- Architecture: MobileNetV2 (Transfer Learning)
- Framework: TensorFlow/Keras
- Validation Accuracy: 63%
- Epochs: 25
Dataset
- Source: Kaggle - niten19/face-shape-dataset
- Classes: Heart, Oval, Round, Square, Oblong
- Training Images: 4000
- Testing Images: 1000
Tech Stack
- Python
- TensorFlow
- MediaPipe
- OpenCV
- Streamlit
ML Pipeline
- Photo Upload
- Face Detection (MediaPipe)
- Face Shape Classification (MobileNetV2)
- Hairstyle Recommendation
- Virtual Try-On Overlay
Evaluation results
- accuracy on face-shape-datasetself-reported0.630