✨ AI StyleMate ✨

AI StyleMate is an AI-powered web service that analyzes a user’s face shape and personal skin tone to provide personalized hairstyle and hair color recommendations. All analysis runs entirely in the browser, ensuring fast performance and strong privacy protection with no image storage.

---

🚀 Live Demo

🔗 https://dadin06-arch.github.io/Pro2/?v=2

No mobile app installation required — the service runs directly in a web browser.

---

🧭 Project Overview

• 🎯 Problem Hairstyles and hair colors are difficult to reverse, yet most people rely on trends or subjective advice when making decisions.

• 💡 Solution AI StyleMate provides data-driven, personalized styling guidance by analyzing the user’s own facial features and skin tone.

• 🔐 Privacy-First Design All AI inference is performed locally in the browser using TensorFlow.js.

---

⚡ Quick Start (Local)

⚠️ No backend server or Python dependencies are required.

1. Run a simple local server

   python -m http.server

2. Open your browser

   http://localhost:8000
   

---

📊 Dataset

• Sources

  • Face Shape: Public datasets from platforms such as Kaggle, combined with limited self-captured experimental data
  • Personal Tone: Public skin tone datasets from open research repositories

• Size: 8,720 images in total

  • Face Shape: 4,000 images (Heart / Oblong / Oval / Round / Square — 800 per class)
  • Personal Tone: 4,720 images (Cool: 2,334 / Warm: 2,386)

• License All datasets are used strictly for academic and non-commercial purposes, in compliance with the original licenses.

---

🧠 Model

• Tool: Google Teachable Machine
• Export Format: TensorFlow.js

📈 Performance

• Face Shape Classification

  • Overall Accuracy: 88%
  • Oval class achieved 100% accuracy

• Personal Tone Classification

  • Accuracy: 88%

• Latency

  • Average inference time: under 0.5 seconds

• Face Detection

  • BlazeFace is integrated to improve real-time face detection and tracking stability

---

🛠️ Workflow

1. 📷 Input The user provides a facial image via webcam or image upload

2. 🧩 Detection Face detection is performed using BlazeFace

3. 🔍 Analysis The system analyzes either face shape or personal tone

4. 🤖 Prediction

   • Face Shape: Heart / Oblong / Oval / Round / Square
   • Personal Tone: Cool / Warm

5. 💇 Suggestion Personalized hairstyle or color palette recommendations are generated

6. 🪞 Preview A real-time AR-style overlay allows users to preview the recommended style

---

🔐 Ethics & Privacy

• User images are never stored or transmitted
• All inference runs locally in the browser
• Dataset limitations and bias were documented and addressed through an Error Board

---

🙌 Credits

• Team: AI StyleMate Group 21 (Kyungmin Lee, Dahye Kim, Hyojin Kim, Kyuwon Lee, Yerim Seok)
• Data: Kaggle, Open Research Skin Tone Datasets
• Tools: TensorFlow.js, Google Teachable Machine, BlazeFace

---

📌 Notes

• This project is intended for academic, non-commercial use only.
• The web interface is fully responsive for both desktop and mobile environments.