|
6DOF VR and 360° Video Datasets
My lab is committed to contributing to community building in many of the
areas we investigate and facilitating follow-up or related work to the studies we carry out.
With these objectives in mind, we have shared publicly multiple valuable VR/360° video
datasets that are described below. If you have any related questions, please do not hesitate
to contact me. If you can, please send us a brief summary sharing how our datasets have benefited your research and the nature of your study. We would be curious and grateful to hear from you. 360° Video Rate-Quality Trade-Offs DatasetDue to the growing presence and importance of emerging perceptual video
quality metrics such as VMAF and SSIM, we have collected a dataset that puts them side by side
with more traditional metrics that have been broadly used in the past such as PSNR and MSE. We
have considered four representative full UHD 360° videos. For each GOP tile of a video, we
have generated and recorded its quality according to these four metrics (PSNR, MSE, SSIM, VMAF),
for different encoding rates. These quality vs. rate trade-offs for the GOP tiles of these four
360° videos have been organized in a dataset, to promote community research and follow up
work.
Please remember to reference the dataset if you use it in your research and paper(s). The dataset should be referenced as: "M. Khan, Z. Xu, and J. Chakareski. NJIT Full UHD (8K) 360° Video Rate-Quality Trade-Offs Dataset. 2024. https://www.jakov.org.", together with the publication below, where it was described and introduced formally, for the first time. J. Chakareski and M. Khan, "Live 360° video streaming to heterogeneous clients in 5G networks," IEEE Trans. Multimedia, Mar. 2024, accepted. NJIT Full UHD 360° Video Rate-Quality Trade-Offs Dataset. Full UHD 360° Video Dataset and Modeling of R-D Characteristics and Head Movement NavigationWe have developed spatiotemporal rate-distortion (R-D) characteristics of
compressed GOP tiles for 15 representative full UHD (8K) 360° video sequences. They
comprise a set of discrete encoding points capturing the rate-distortion trade-offs of
a tile (i,j) in the 360° equirectangular panorama over the duration of a GOP. We have
also pursued accurate analytical modeling of these discrete points to functionally characterize
the rate-distortion dependencies they capture.
We have recorded corresponding head movement traces describing how users wearing a virtual reality headset have navigated these 360° videos on the headset during a display session. For each session (user-video pair), these data capture the head orientation of the user in the form of three canonical angles (yaw, pitch, and roll) referenced on the headset, for every temporal video frame of the 360° content displayed on the headset worn by the user during the session. The head movement data is accompanied by statistical modeling of the navigation actions of a user across the tiles of the 360° panorama for the duration of a GOP, for every video in the dataset. The developed dataset and modeling can be beneficial in key application cases such as streaming system rate allocation, distribution system caching and transcoding, and perceptual studies and immersion saliency. Please remember to reference the dataset if you use it in your research and paper(s). The dataset should be referenced as: "J. Chakareski and R. Aksu. Full UHD 360° Video Dataset and Modeling of R-D Characteristics and Head Movement Navigation. 2021. https://www.jakov.org.", together with the publication below, where it was described and introduced formally, for the first time. J. Chakareski, R. Aksu, V. Swaminathan, and M. Zink, "Full UHD 360° video dataset and modeling of rate-distortion characteristics and head movement navigation," in Proc. ACM Multimedia Systems Conference, Istanbul, Turkey, Sept. 2021. Full UHD 360° Video Dataset and Modeling. NJIT 6DOF VR Navigation DatasetWe captured 6DOF (six degrees of freedom) virtual reality (VR)
navigation data comprising spatial position (x,y,z) and head orientation (rotation
angles yaw, pitch, and roll) of mobile VR users navigating a VR environment in an
indoor arena. We share this dataset publicly here to motivate investigations of future
mobile wireless network systems and a broader community engagement. The dataset can
also facilitate diverse studies that explore the perceptual interaction and quality
of experience of users in 6DOF immersive environments.
In our setup, the users were navigating the 6DOF VR content Virtual Museum across a spatial area of 6m x 4m, divided into six playing areas (cells) of size 2m x 2m each (height is 3m). The users were wearing the HTC Vive Wireless VR headset (HMD). We captured data for three volunteer users individually, across six sessions per user, one for each cell (its center) used as the starting navigation point for the user. 30,000 tracking samples were captured per session, at 250 samples per second. For further details, please refer to one of the publication(s) below. Please remember to reference the dataset if you use it in your research and paper(s). The dataset should be referenced as: "M. Khan and J. Chakareski. NJIT 6DOF VR Navigation Dataset. 2020. https://www.jakov.org.", together with one of the publication(s) below, where it was described and introduced formally, for the first time. J. Chakareski and M. Khan, "Wifi-VLC dual connectivity streaming system for 6DOF multi-user virtual reality", in Proc. ACM Int'l Workshop on Network and Operating Systems Support for Digital Audio and Video (NOSSDAV), Istanbul, Turkey, Sep. 2021. J. Chakareski, M. Khan, T. Ropitault, and S. Blandino, "6DOF Virtual Reality Dataset and Performance Evaluation of Millimeter Wave vs. Free-Space-Optical Indoor Communications Systems for Lifelike Mobile VR Streaming", in Proc. IEEE Asilomar Conference on Signals, Systems, and Computers, Pacific Grove, CA, USA, Nov. 2020. NJIT 6DOF VR Navigation Dataset. |