Smart Hairstyle

A hairstyle recommendation and try-on app based on our CelebHair Paper, supervised by Prof. Jinpeng Chen.

• Led a team of four in developing a hairstyle recommendation system that suggests optimal hairstyles based on facial images. Further introduced a try-on feature, aiding hairstylists in visualizing recommended hairstyles, addressing a persistent industry challenge
• Introduced CelebHair, a pioneering large-scale dataset for hairstyle recommendation, setting a new benchmark in terms of variety, veracity, and volume compared with existing hairstyle-related datasets
• Enhanced face shape classification performance using mosaic data augmentation, batch size auto-calculation, and CIoU. Employed YOLO with darknet as the backbone, achieving an 87.45% accuracy across five face shapes — a 15% advancement over leading existing methods
• Utilized Spatial Transformer Network for hairstyle classification, achieving enhanced robustness by learning invariance to image translations, scaling, and rotations, crucial for diverse hairstyle shapes and textures
• Developed a hairstyle recommendation system using Random Forests with scikit-learn, ingeniously transforming the recommendation problem into a classification task, attaining an impressive 87.03% accuracy
• Devised a face-swapping algorithm using OpenCV and facial key points matching, creating a ”Virtual Mirror” feature, empowering users to virtually experiment with various hairstyles
• Crafted a hairstyle recommendation application using Vue, Flask and SQLite, designed to assist hairstylists. This app notably enhanced service quality, resulting in improved satisfaction levels within collaborative barbershops