Shoppin 🍓

Shoppin is an AI-powered fashion discovery platform that transforms visual inspiration into shoppable products. It combines the browsing experience of Pinterest with intelligent, computer-vision-powered product matching.

---

🚀 Quick Start

Run the entire stack (Frontend, Backend, Database) with a single command:

docker compose up --build

The application will be available at:

• Frontend: http://localhost:3000
• Backend API: http://localhost:8000

---

💡 What Problem Does It Solve?

Have you ever seen an amazing outfit online and wondered, "Where can I buy that?"

Traditional search engines fail here because fashion is visual. Describing a specific "vintage oversized beige trench coat" with keywords is difficult and often inaccurate.

Shoppin solves this using AI. We use computer vision to "see" the outfit, understand the style, and instantly find similar products you can actually buy.

✨ Key Features & How They Work

1. Shop the Look (YOLOv8 + Intelligent Cropping)

Click on any outfit, and Shoppin identifies the individual items (shoes, tops, pants) and finds matches.

How it works:

1. Detection: We use YOLOv8-Pose to detect the person and 17 body keypoints (shoulders, knees, ankles, etc.).
2. Precision Cropping: Instead of generic bounding boxes, we use keypoints to crop exact regions (e.g., "Ankle to Floor" for shoes).
3. Matching: The cropped region is converted to a vector embedding and matched against our product catalog.

2. Visual Search (CLIP Embeddings)

Upload any image to find visually similar products. You can refine results with text ("no leather") or price filters.

The Tech Stack:

• CLIP Model: Converts images into 512-dimensional vector embeddings.
• pgvector: Performs cosine similarity search in PostgreSQL to find the closest visual matches in milliseconds.
• Hybrid Search: Combines visual embeddings with text embeddings for refined queries.

3. Semantic Shop Search (Spell Tolerant)

Search for products using natural language, even with typos or slang.

Example:

• Query: "micheal baskelball shoes" (Typos included)
• Result: Michael Jordan Basketball Shoes

Why? We use Sentence Transformers to match the meaning (semantics) of your query rather than exact keywords. The vector for "micheal" is nearly identical to "michael," so the search just works.

---

🛠️ Technology Stack

• Frontend: Next.js, React, TailwindCSS, Lucide Icons
• Backend: Django, DRF, Gunicorn
• Database: PostgreSQL, pgvector
• AI Models:
  • yolov8n-pose.pt (Person & Keypoint Detection)
  • CLIP (Visual Embeddings)
  • all-MiniLM-L6-v2 (Text Embeddings)
• Infrastructure: Docker, Docker Compose

Architecture (Below is the asynchronous architecture design. The MVP implementation is synchronous, but the system is designed to scale using Celery workers, Redis queues, and asynchronous job processing.)

Shiping Live Demo:-

Shoppin.1.mov

📝 License

MIT