STAV360: A Dataset for Subjective Tile-based Assessment of 360° Videos

Tile-based 360° video streaming has recently emerged as a promising strategy for efficient and flexible 360° video transmission. However, spatial variations in the delivered scene quality, caused by combining different quality levels, can influence the users' perceptual experience. While several models and datasets have been developed for 360° video quality assessment (360° VQA), none of the available studies consider the spatial quality variation that arises in multi-quality tiled 360° video streaming. To bridge this gap, conducting subjective evaluation experiments for tiled omnidirectional videos is essential for developing reliable tiled 360° VQA models and optimized streaming algorithms. This paper introduces the STAV360 dataset, a dataset for subjective tile-based assessment of 360° videos. The STAV360 dataset consists of six 360° video sequences, each tiled and encoded using twelve different patterns, along with subjective ratings and tracking data of 27 participants. STAV360 reveals interesting patterns in subjective tiled 360° VQA, paving the way for new research opportunities in tile-based 360° video streaming and objective 360° VQA metric models. The STAV360 dataset is publicly available at https://github.com/SLG-European-Projects/STAV360.