Inhaltszusammenfassung

Selecting suitable 3D datasets for specific tasks, especially in machine learning, remains a challenge due to the growing number of datasets and the lack of uniformly comparable selection criteria. This paper addresses this gap by analyzing how application-relevant characteristics are distributed across 263 publicly available 3D datasets, including widely used datasets such as KITTI and ScanNet. We examine five exemplary characteristics to capture aspects such as temporal distribution, public...Selecting suitable 3D datasets for specific tasks, especially in machine learning, remains a challenge due to the growing number of datasets and the lack of uniformly comparable selection criteria. This paper addresses this gap by analyzing how application-relevant characteristics are distributed across 263 publicly available 3D datasets, including widely used datasets such as KITTI and ScanNet. We examine five exemplary characteristics to capture aspects such as temporal distribution, publication venues, benchmark integration, academic impact, and scene type. The results reveal both an increase in dataset releases in recent years and notable imbalances in the current dataset landscape. To support transparent dataset selection and future research, we provide a curated and openly accessible GitHub repository.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
4.43 - Informatik

DDC Sachgruppe:
Informatik