🕹️ Synthetic data generator

📁 Table of Contents

• General
• Requirements
  • Collections and plugins
  • Expected tree
• Configuration
• Contributors

🏙️ General

This repository generates synthetic data of Paris from OpenStreetMap data and animates it with the City Sample project (pedestrians, cars). The synthetic data is captured in several formats (RGB and semantic).

The objective is to detect AI models biases from this synthetic data. For instance, detect if our crowd counting model is efficient either day and night by submitting two datasets.

🧰 Requirements

• Unreal Engine 5.1.1

Collections and plugins

In your project, you will need to add:

• City Sample Crowds
• City Sample Vehicles

Expected tree

. [Twincity]
└── Content # Download from the marketplace 👇
  └── City Sample Crowds
  └── City Sample Vehicles
├── [...]
└── Plugins # Create a Plugins folder where you can clone EasySynth repository 👇
    └── EasySynth

💾 Demo

🌙☀️ Day and night scenes

🌧 Weather

📸 Camera angles and semantic views

Semantic views is useful in computer vision to detect objects (pedestrians, buildings, roads, vehicles etc).

👥 Crowd generator

📊 Extractor results

Once the user has clicked on "Generate", RGB and semantic datas can be found in /d/Snapchot/ folder. Metadatas file describing screenshots (number of vehicles and peds, peds id and RGB, camera rotation/location, weather...) is also created in /d/.

👋 Contributors

• Lead Dev - Jehanne Dussert
• Dev - Rémi Giner
• Dev - Nobila Traore
• Game Artist - William Lahemar