ToF-360 Panoramic RGB-D Dataset for 3D Reconstruction

This technical report presents the ToF-360 dataset, a novel RGB-D dataset designed for indoor semantic 3D reconstruction. The dataset is captured using a unique Time-of-Flight sensor that allows for 360-degree omnidirectional scanning within seconds, addressing the limitations of existing RGB-D datasets that require multiple scans and often result in stitching artifacts and low-quality annotations. The ToF-360 dataset includes fisheye and ERP images, along with high-precision pixel-level 2D semantics and room layout annotations. It introduces benchmarks for 2D semantic segmentation, 3D semantic segmentation, and layout estimation, enhancing the representation of real-world environments and improving the performance of existing methods. The report details the dataset's creation process, its advantages over traditional datasets, and its potential applications in various fields such as augmented reality, autonomous driving, and robotics. The dataset is publicly available for further research and development in scene understanding tasks.