🧵 Second Stitch

AI-Powered Sustainable Wardrobe Upcycling System

Upload a photo of any garment → AI classifies it → automatically transforms it into an upcycled version → view side-by-side before/after comparison.

---

🌿 What is Second Stitch?

Second Stitch is a full-stack AI application that takes photos of second-hand or existing clothing and automatically upcycles them into new, modern garments — reducing textile waste through intelligent automation.

The system combines:

• Deep learning classification (EfficientNet-B3, 91.45% accuracy) to identify the garment type
• Semantic segmentation (SegFormer-B3) to isolate garment pixels and erase the person/background
• Deterministic geometric transformation to physically shorten hems and remove/shorten sleeves
• Generative AI finishing (SDXL Inpainting) to produce a clean, sewn-looking edge at the cut

---

✨ Features

• 📸 Upload any clothing photo — product shots, worn photos, flat lays
• 🤖 Automatic garment type detection across 13 categories
• ✂️ Smart upcycling: long dress → mini dress, trousers → shorts, long sleeve → crop top, etc.
• 🖼️ Interactive before/after slider comparison
• 💾 Save, favourite, and share your upcycled designs
• 🔐 User authentication via Supabase
• ⚡ FastAPI backend with GPU acceleration

---

🗂️ Project Structure

Second_Stitch/
├── backend/
│   ├── main.py                  # FastAPI app entry point
│   ├── models/
│   │   ├── classifier.py        # EfficientNet-B3 garment classifier
│   │   └── sd_pipeline.py       # SegFormer + geometric transform + SDXL pipeline
│   ├── services/
│   │   ├── detection.py         # Image validation
│   │   ├── generation.py        # Orchestration: classify → transform → save
│   │   └── prompt_engine.py     # Upcycle rules + SDXL fabric prompts
│   ├── routes/
│   │   └── upcycle.py           # POST /api/upcycle endpoint
│   └── uploads/                 # Generated images served statically
├── frontend/
│   ├── src/
│   │   ├── pages/               # Route page components
│   │   ├── components/          # Reusable UI components
│   │   ├── lib/                 # Supabase client config
│   │   └── main.jsx             # Entry point
│   ├── public/
│   ├── vite.config.js
│   └── package.json
├── checkpoints/
│   └── best.pt                  # Trained EfficientNet-B3 checkpoint
├── scripts/
│   ├── train.py                 # Model training script
│   └── inference.py             # Standalone inference script
└── docs/
    ├── ARCHITECTURE.md
    ├── TRAINING.md
    ├── API.md
    └── DEVELOPMENT_JOURNEY.md

---

🚀 Quick Start

Prerequisites

• Python 3.12+
• Node.js 20+
• CUDA-capable GPU (recommended: 16GB+ VRAM)
• Supabase account

1. Clone the repository

git clone -b Final_Project https://github.com/HarshitaSoni24/second-stitch.git
cd second-stitch

2. Backend setup

# Install Python dependencies
pip install -r requirements.txt

# Models download automatically on first run (~8GB):
# - sayeed99/segformer-b3-fashion
# - diffusers/stable-diffusion-xl-1.0-inpainting-0.1

# Start the backend
cd Second_Stitch
uvicorn backend.main:app --host 0.0.0.0 --port 8080

3. Frontend setup

cd frontend

# Install dependencies
npm install

# Create environment file
echo "VITE_API_URL=http://localhost:8080" > .env

# Start development server
npm run dev

4. Open in browser

http://localhost:5173

---

🧠 Model Performance

---

🔄 Upcycle Logic

---

🛠️ Tech Stack

Backend: Python 3.12, FastAPI, PyTorch 2.5.1, CUDA 12.2, timm, HuggingFace Transformers, Diffusers, OpenCV, rembg, Pillow

Frontend: React 18, Vite 7, Tailwind CSS, Framer Motion, React Router DOM, lucide-react, html2canvas

Database & Auth: Supabase (PostgreSQL + JWT Auth)

Infrastructure: NVIDIA RTX A6000, JupyterHub GPU Server, ngrok

---

👩‍💻 Team

Guided by: Dr. P. Kayal, Professor & HoD, Department of IT
Institution: BVRIT HYDERABAD College of Engineering for Women

---

📄 License

This project is developed for academic purposes at BVRIT Hyderabad.

---

📚 Documentation

• Architecture Overview
• Model Training Guide
• API Reference
• Development Journey