VideoSet, short for Video Subject Evaluation Test, is a large-scale compressed video quality dataset based on the Just Noticeable Difference (JND) technique. The dataset consists of 220 5-second video sequences, each with four resolutions (i.e., 1,920×1,080, 1,280×720, 960×540, and 640×360), for a total of 880 videos. The researchers used the H.264 codec to encode each video with QP=1,⋯,51, and measured the first three JND points with more than 30 subjects.

The source code/encoding, video clips, and measured JND data involved in this dataset are publicly available, which is of great significance to video coding research and standardization work.