GARBGENIUS: AR APPROACH ENHANCING WARDROBE EXPERIENCE WITH RECOMMENDATION SYSTEM

Overview

GarbGenius is an AI-driven fashion recommendation system that enhances the shopping experience with Augmented Reality (AR) and machine learning algorithms. It allows users to virtually try on clothes and receive personalized fashion recommendations based on their preferences and current trends. Initially, K-Nearest Neighbors (KNN) was used for recommendations, but Support Vector Machine (SVM) was later integrated to improve accuracy. (I pulled recommendation system source code only, If you want AR code ask me at gmail : [email protected]), Here publication for more detail [https://ijcrt.org/viewfull.php?&p_id=IJCRT24A4814]

Features

• AI-Powered Clothing Recommendations
• Virtual Try-On with Augmented Reality (AR)
• Vision Transformer (ViT) for Image Feature Extraction
• SVM for Enhanced Recommendation Accuracy
• Graphical User Interface (Tkinter-based)

Technologies Used

• Python, Flask (Backend Development)
• Vision Transformer (ViT) (Feature Extraction)
• Support Vector Machine (SVM) (Classification & Recommendations)
• Pandas, NumPy (Data Handling)
• Tkinter (GUI for User Interaction)
• OpenCV & PIL (Image Processing)
• MeshLab (3D Visualization)
• Joblib (Model Storage & Loading)

Algorithmic Workflow

1. Image Feature Extraction

• Uses Vision Transformer (ViT) to extract clothing features from user-uploaded images.
• Converts images into numerical feature representations.

2. Recommendation System

• KNN was initially used but later replaced with SVM, which improved accuracy.
• The model is trained on labeled clothing datasets to predict recommendations.
• Uses Cosine Similarity to match user preferences.

3. Virtual Try-On Experience

• Integrates Augmented Reality (AR) to allow users to visualize how recommended outfits would look on them.
• Uses OpenCV and MeshLab for visualization.

Performance & Accuracy

Model	Accuracy
ResNet50 + KNN	75%
ViT + Cosine Similarity	77.9%
ViT + SVM (Final Model)	80.8%

The SVM-based approach provided the highest accuracy, leading to its adoption.

Future Improvements

• Expand dataset for better recommendations.
• Implement Deep Learning-based virtual try-on.
• Deploy a web-based interactive UI.

License

This project is MIT licensed. Feel free to contribute!

Contributing

Contributions are welcome! Open an issue or submit a pull request.

---

Author

Santhosh D Suryaa narayanan K Thiyaneshwaran S Vishal ponn rangan K