Fashion Outfit Recommendation System

Welcome to the ** Fashion Outfit Recommendation System** dataset! This package contains a curated subset of fashion items and pre-styled outfits designed to evaluate your capabilities in computer vision, recommendation systems, search retrieval, and natural language understanding.

---

📁 Dataset Directory Structure

This curated dataset is organized as follows:

NEWDATASET/
├── README.md               # This documentation file
├── curated25.xlsx          # Original styled outfits spreadsheet
├── outfits.csv             # Cleaned outfit mapping of the 25 curated outfits
├── products.csv            # Detailed metadata for the 68 unique products used in the outfits
└── images/                 # Product image files matching the product IDs
    ├── ajio/               # Images sourced from Ajio
    ├── myntra/             # Images sourced from Myntra
    └── nykaa/              # Images sourced from Nykaa

---

📊 Data Files Description

1. products.csv

This file contains the core metadata for the 68 unique fashion items that make up our outfits.

• Fields:
  • id: Unique identifier for the product (e.g., ajio_703182002).
  • name: Product title (e.g., Women Bodycon Midi Length Dress).
  • brand: Manufacturer or label (e.g., Fyre Rose, Peter England).
  • price_inr: Retail price in Indian Rupees (INR).
  • rating / rating_count: Customer rating statistics.
  • gender: Target gender (men / women).
  • wear_type: Style category (e.g., western, ethnic).
  • category & category_label: Specific clothing/accessory category (e.g., formal-shirts, heels, dresses).
  • occasion: Intended setting (e.g., party, office, casual).
  • tags: Semicolon-separated tags for retrieval.
  • description: Detailed text description of the product.
  • image: Relative filepath to the product image (e.g., images/ajio/703182002.jpg).

2. outfits.csv (and curated25.xlsx)

This file defines 25 expert-curated complete outfits. You can use this file as ground truth for training, evaluation, or as reference combinations.

• Fields:
  • outfit_id: Unique identifier for the outfit (e.g., outfit W1).
  • gender / wear_type / occasion / theme: Categorization context.
  • hero & hero_id: The main item in the outfit (e.g., a dress or shirt).
  • second & second_id: The complementary item (e.g., trousers/chinos).
  • layer & layer_id: Optional layering item (e.g., blazers, jackets).
  • footwear & footwear_id: Footwear item.
  • accessory_1 & accessory_1_id / accessory_2 & accessory_2_id: Optional styling accessories.
  • palette: Main color combination.
  • stylist_rationale: Stylist commentary explaining why this outfit is compatible and fits the theme.

---

🎯 Assignment & Problem Statement

Your objective is to design and build an intelligent Recommendation Engine & Chat-based Fashion Assistant that can understand natural language user requests, retrieve compatible clothing items, compile complete outfits, and explain its reasoning.

Core Implementation Requirements:

1. Dataset Analysis & Understanding:

   • Inspect the provided metadata and images.
   • Document categories, palette distributions, and potential challenges (e.g., size of dataset, variety, metadata consistency).

2. Outfit Compatibility Engine:

   • Build an algorithm to determine if items are compatible. Given a single item (e.g., a white formal shirt), the engine should suggest compatible components (e.g., navy trousers and brown loafers).
   • Tip: Use similarity search or learn a pairwise compatibility score using visual/text features.

3. User & Context-Aware Recommendations:

   • Adapt recommendations dynamically based on profile parameters:
     • Gender (e.g., Men / Women)
     • Age Group (e.g., 20s vs. 40s styling)
     • Occasion (e.g., Office, Beach Vacation, Wedding, Party)
     • Style Preferences (e.g., Formal, Smart Casual)

4. Conversational Fashion Assistant (Natural Language Interface):

   • Build a chat interface allowing users to make requests in plain text (e.g., "I need an outfit for a business meeting" or "Suggest something stylish for a summer beach vacation").
   • The assistant should retrieve the items, group them into a complete outfit (Topwear, Bottomwear, Footwear, and optional Accessories/Layers), and display them to the user.

5. Explainability:

   • Every outfit recommendation must include a reasoned explanation (e.g., "Beige chinos pair well with a navy blazer because they provide classic contrast while maintaining a polished smart-casual appearance for your office meeting.").

---

🛠️ Recommended Technical Approach

We evaluate technical depth and systemic engineering choices. Consider incorporating:

• Computer Vision & Multi-modal Embeddings: Use models like CLIP, FashionCLIP, or SigLIP to generate embeddings from both product images and descriptions.
• Vector Search / Retrieval: Store product embeddings in a vector database (e.g., Qdrant, Chroma, FAISS) to execute fast similarity and hybrid searches.
• LLM Integration: Use LLMs (e.g., Gemini, GPT, Claude) to parse user intent from conversational chat, structure queries, and generate final personalized reasoning.
• Advanced Methods (Bonus): Graph-based recommendations (representing outfits as nodes/edges) or trained compatibility classification models.

---

🚀 Quick Start Code (Python)

You can load and start exploring this dataset using the following snippet:

import pandas as pd
import os

# Set paths
DATASET_DIR = "./"  # Update path if run from elsewhere
products_df = pd.read_csv(os.path.join(DATASET_DIR, "products.csv"))
outfits_df = pd.read_csv(os.path.join(DATASET_DIR, "outfits.csv"))

print(f"Loaded {len(products_df)} products.")
print(f"Loaded {len(outfits_df)} curated outfits.")

# Example: Display first outfit
first_outfit = outfits_df.iloc[0]
print(f"\nOutfit ID: {first_outfit['outfit_id']} ({first_outfit['theme']})")
print(f"Hero Item: {first_outfit['hero']} (ID: {first_outfit['hero_id']})")
print(f"Footwear: {first_outfit['footwear']} (ID: {first_outfit['footwear_id']})")
print(f"Rationale: {first_outfit['stylist_rationale']}")