README.md

Fashion Product Recommendation System 🛍️

This project demonstrates a content-based fashion product recommender system with advanced capabilities, including outfit compatibility evaluation, image-based recommendations, and product filtering. The application integrates cutting-edge machine learning techniques to provide an intuitive and fast recommendation experience.

Features 🌟

1. Product Search and Filtering

   • Utilizes TF-IDF to preprocess and classify over 21,000 apparel items across 57 unique categories.
   • Includes filters for categories, price range, and product attributes for refined searches.

2. Outfit Compatibility Recommender

   • Integrates the Google Gemini API to analyze clothing attributes and recommend complementary outfit items.

3. Image-Based Recommendation

   • Fine-tunes the ResNet50 model to extract image embeddings for precise similarity-based recommendations.
   • FAISS ANN (Approximate Nearest Neighbors) integration reduces recommendation latency to 0.013 seconds, achieving a 16.6x speed improvement over traditional cosine similarity.

Results 📊

• Classification Accuracy: Fine-tuning ResNet50 resulted in a 30% improvement (from 43.8% to 73.0%) in classification accuracy on test data.
• Latency Optimization: FAISS-powered recommendations achieved real-time performance, making the system highly scalable.

Demo 🌐

The application is deployed on Streamlit and offers:

1. Product Search: Search by keywords or filter by category, price, and other attributes.
2. Outfit Combination Recommender: Suggests complementary items for a given clothing piece.
3. Image-Based Recommendation: Upload an image to find visually similar apparel items.

How to Use 🚀

Prerequisites

• Install Python 3.8 or above
• Clone this repository and navigate to the project folder
• Install dependencies using:

  pip install -r requirements.txt  

Run the App

1. Launch the Streamlit app:

   streamlit run app.py  

2. Upload an image or search for products to explore the system's features.

Sample Data

The project uses a filtered dataset of 21,000+ apparel items. Ensure the dataset and required pre-trained model files (image_embeddings1.npy and final_articleType_model1.h5) are present in the project directory.

Repository Structure 📂

• app.py: Main Python script integrating all features and the Streamlit app.
• data/: Dataset folder (21k+ apparel items).
• models/: Pre-trained model files and embeddings.
• requirements.txt: List of Python dependencies.

Technical Details 🔍

• Content-Based Filtering: TF-IDF vectorization for text-based product search.
• ResNet50: Fine-tuned for image classification and feature extraction.
• FAISS ANN: Reduces latency for large-scale image similarity search.
• Google Gemini API: Analyzes attributes to recommend complementary outfits.