"University of Aizu Neural Rendering (UoANeR) Dataset"

"Novel view synthesis (NVS) aims to reconstruct a scene representation from limited observations and render previously unseen viewpoints. It has become a foundational component of computer vision and computer graphics systems with the advancement of recent neural rendering methods such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Within the various  NVS applications, a quality evaluation dataset plays a critical role by supporting the quality evaluation of models. However, existing neural rendering quality evaluation datasets lack diversity in both scene content and rendering models, limiting their effectiveness and generalizability. The UoANeR dataset is a large-scale collection of rendered images (approximately  20K) from over 50 diverse scenes, generated using both NeRF-based and 3DGS-based models within a controlled rendering pipeline. It provides critical support for quality evaluation, benchmarking of quality metrics, comparative analysis of model performance, and distortion identification in neural rendering. All data are stored in standard, easily accessible file formats."