360-Degree Video Experiment Dataset

To study the viewer motion with head-mounted displays (HMDs), we setup an experimental environment with Oculus DK2. We developed a software to play 16 videos automatically to viewers. After each view of a video clip, the subject is required to provide a subjective score in the range of 1 to 5. This score is used to detect whether the video content is interesting to the viewer or not. On the other hand, it helps the subject to concentrate on the whole video clip. When a subject is watching a video, his or her motion is recorded and logged. The motion includes 3 degrees of freedom, pitch, yaw, and roll (i.e., X, Y, and Z angles). When wearing a HMD, the viewer's initial position defines the zero degree for pitch, yaw, and roll. Each dimension is denoted by an angle ( -180 degree to 180 degree).

We use Oculus DK2 as the hardware, Oculus runtime 7.0 as the hardware driver and Color Eyes as the video player. Subjects are sitting on a chair that can rotate horizontally in 360 degree. The HMD of Oculus DK2 is connected to a PC with a cable.

In total, 153 volunteers joined the experiment, where 35 of them watched all 16 video clips and 118 of them watched 3~5 randomly-selected video clips. Most of the volunteers are people on campus, and most of them have their first VR experience in our experiment. The age distribution is as follows: 10~20 (36%), 20~30 (54%), 30~40 (4%), 40~50 (4%), 50~60 (2%). 38% of the volunteers are female, and 34% of the volunteers wear glasses.

We download 16 clips of 360-degree video from Youtube and cut each of them into 30 seconds. We manually choose the highest resolution available on Youtube for these videos. Among the 16 videos, 14 are of 4K resolution, one is of 2K resolution, and the other one is of 1080P. All the videos have 30 frames per second. As a first attempt, we remove the audio of the video clips, such that subjects' motion is only related to visual content. Each subject is asked to watch the video clips in order, meanwhile the motion of head is recorded. 7-9 measurements of X, Y and Z angles are recorded per second.
The full dataset could be downloaded here. More information could be found in our paper (IEEE BigData 2016) [Shooting a Moving Target: Motion-Prediction-Based Transmission for 360-Degree Videos] and poster (ICNP 2016) [Viewing 360 Degree Videos: Motion Prediction and Bandwidth Optimization]. To illustrate the format of data we collected, we provide a sample file which shows one sample viewer motion trace (30s) for one video here. The sample file includes 4 columns: 1st columns is the time (0-30s); the 2nd to the 4th columns correspond to the recorded Y, X and Z angles ( -180 degree to 180 degree), respectively.