Welcome to CarlaOcc Dataset Documentation!

Homepage   |   Video   |   Paper   |   Dataset   |   GitHub

CarlaOcc is a high-fidelity and physically-consistent dataset created from the CARLA simulator. To facilitate the development of 3D perception, scene understanding, and occupancy prediction tasks, CarlaOcc provides comprehensive multi-modal data and fine-grained annotations, including:

• Surround-view/Stereo RGB images

• Panoptic/semantic occupancy ground truth

• Depth images

• Surface normal images

• Semantic segmentation images

• LiDAR/Semantic LiDAR point clouds

• Traffic metadata (e.g., 3D bounding boxes, vehicle velocities)

• Camera/LiDAR poses

• Static meshes of the scene

• Background scene layouts

This project provides complete functionality for data collection, scene exportation, and occupancy ground truth generation.

Getting Started

We recommend the readers to explore this project by the following sequences:

• Dataset Tutorial (CarlaOcc/tutorials/):

  Have a quick look of our dataset

• Data Collection (CarlaOcc/data_collection/):

  Collect sensor data from the CARLA simulator

• Scene Exportation (CarlaOcc/scene_exportation/):

  Export scene geometry layout and semantic information

• Occupancy Grid Generation (CarlaOcc/occupancy_generation/):

  Generate dense and sparse occupancy grid representations

Documentation

• Dataset Tutorial
  • Prepare Environment
  • Prepare Dataset
  • Dataset Visualization
  • Coordinate system definition
• Data Collection
  • Overview
  • Prerequisites
  • Sensor configuration
  • Running
  • Output Data Format
• Scene Exportation
  • Overview
  • Running
  • Output Data Format
• Occupancy Generation
  • Overview
  • Running
  • Occupancy Resolution Settings
  • Panoptic Occupancy Labels

Note

This project is under active development. Feel free to contribute to the project by opening an issue or a pull request.