Realistic benchmark RBD360 dataset for quality assessment of random user generated 360° videos

User-generated 360° video content is becoming increasingly popular due to the availability of low-cost acquisition devices and its inherently immersive nature. To facilitate research and development in this area, we introduce RBD-360, a dataset designed explicitly for quality assessment of user-generated 360° videos that includes various distortions typical of non-professional settings. RBD-360 features a collection of randomly captured source videos across multiple categories, recorded in three different environments: indoor, outdoor during the day, and outdoor at night. The impaired video sequences are created using three standard codecs-H.264, H.265, and VP9 at varying bitrates and quantization parameters. To evaluate subjective video quality, we employed an HMD-based Modified Absolute Category Rating protocol, along with assessments from the Simulator Sickness Questionnaire. This helps capture time-dependent viewer discomfort according to ITU-T P.919 recommendations. We implement benchmark quality assessment metrics on the RBD-360 dataset and report the results. The dataset is publicly available for testing quality assessment models tailored for user-generated videos at https://github.com/manav2701/RBD-360.