This dataset was proposed by the University of Queensland, the University of Technology Sydney, and Ideal Auto in 2024. It is the first large-scale 3D real car dataset, containing 2,500 cars photographed in real scenes.3DRealCar: An In-the-wild RGB-D Car Dataset with 360-degree Views".

3DRealCar aims to be a valuable resource to facilitate car-related tasks. 3DRealcar contains detailed annotations of various colors, car types, brands, and even car parsing diagrams. In particular, the dataset contains three lighting conditions of car surfaces and poses challenges to existing methods.

Dataset characteristics

• High capacity: 2,500 cars were meticulously scanned by 3D scanners to obtain car images and point clouds with real-world dimensions;
• High quality: Each car is captured in an average of 200 dense, high-resolution 360-degree RGB-D views, enabling high-fidelity 3D reconstruction;
• High Diversity: The dataset contains a variety of cars from more than 100 brands, collected under three different lighting conditions including reflective, standard, and dark.