Interia - AI-Powered Room Redesign

Transform any room with AI-powered 3D reconstruction and intelligent furniture replacement

Demo

---

Overview

Interia is a full-stack AI application that transforms video recordings of rooms into interactive 3D point cloud visualizations. Upload a video of your room, and our AI pipeline powered by Depth Anything V3 reconstructs it in 3D, detects furniture using YOLOv8, and enables intelligent furniture replacement with Google Gemini image generation.

Features

🎥 Video-to-3D Reconstruction

• Upload room videos with drag-and-drop support
• Real-time depth estimation using Depth Anything V3
• Multi-view depth with spatial consistency
• Interactive 3D point cloud visualization with Three.js
• Camera pose estimation (intrinsics & extrinsics)

🛋️ AI Furniture Detection & Search

• YOLOv8-powered furniture detection from room screenshots
• Semantic product search with sentence transformers
• Sponsored brand boosting system
• Product catalog with 45,000+ items

🎨 Smart Furniture Replacement

• AI-powered furniture replacement with Google Gemini
• Ultra-realistic photorealistic rendering
• LRU caching with TTL for generated images
• Preview replacements before committing

Tech Stack

Frontend

Technology	Purpose
React 18 + TypeScript	UI Framework
Vite 7	Build Tool
Three.js	3D Visualization
Tailwind CSS	Styling
React Router	Navigation

Backend

Technology	Purpose
FastAPI + Uvicorn	API Server
Depth Anything V3	Depth Estimation
YOLOv8	Object Detection
Google Gemini	AI Image Generation
Sentence Transformers	Semantic Search
SQLAlchemy + SQLite	Database
Docker + NVIDIA CUDA	GPU Acceleration

Quick Start

Prerequisites

• Node.js 18+ / Bun (Frontend)
• Python 3.11+ (Backend)
• NVIDIA GPU with CUDA 12.1+ (Required for AI processing)
• Docker with nvidia-container-toolkit (Recommended)

Frontend Setup

# Install dependencies
bun install

# Start development server
bun run dev

The frontend runs at http://localhost:5173

Backend Setup

Option 1: Docker (Recommended)

cd backend

# Production
docker-compose up --build

# Development (with hot reload)
docker-compose -f docker-compose.dev.yml up --build

Option 2: Local Python

cd backend

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/macOS
# or: .\venv\Scripts\activate  # Windows

# Install dependencies
pip install -r requirements.txt
pip install git+https://github.com/ByteDance-Seed/Depth-Anything-3.git

# Start server
uvicorn app.main:app --reload --port 8000

The backend runs at http://localhost:8000

Project Structure

interia/
├── src/                          # Frontend (React)
│   ├── components/               # React components
│   │   ├── VideoUpload.tsx       # Video upload with drag-drop
│   │   ├── ProcessingStatus.tsx  # Progress tracking
│   │   ├── PointCloudViewer.tsx  # Three.js 3D viewer
│   │   ├── FurnitureSearch.tsx   # Product search UI
│   │   └── ...
│   ├── pages/                    # Route pages
│   │   ├── Home.tsx              # Main app page
│   │   ├── MyRooms.tsx           # Saved rooms gallery
│   │   ├── RoomViewer.tsx        # Individual room view
│   │   └── ...
│   ├── contexts/                 # React contexts
│   └── services/                 # API clients
├── backend/                      # Backend (FastAPI)
│   ├── app/
│   │   ├── api/                  # API routes
│   │   ├── services/             # Business logic
│   │   ├── db/                   # Database models
│   │   └── models/               # Pydantic schemas
│   ├── Dockerfile                # GPU-enabled container
│   └── docker-compose.yml        # Production config
├── data/                         # Product catalog CSV
└── public/                       # Static assets

Configuration

Environment Variables

Variable	Default	Description
GARAZA_MAX_FRAMES	16	Frames to extract from video
GARAZA_MODEL_NAME	da3-base	DA3 model variant
SECRET_KEY	-	JWT signing key
GOOGLE_API_KEY	-	Google Gemini API key

DA3 Model Variants

Model	VRAM	Speed	Quality
da3-small	2GB	Fastest	Good
da3-base	4GB	Balanced	Better
da3-large	8GB+	Slower	Best

Acknowledgments

• Depth Anything V3 by ByteDance
• YOLOv8 by Ultralytics
• Three.js for 3D visualization
• Google Gemini for AI image generation

---

Built with ❤️ for the future of interior design