The data can be obtained at: https://olimar.stanford.edu/hdd/lvsdataset/

The Long Video Streams (LVS) Dataset was introducted in the paper Online Model Distillation for Efficient Video Inference by Mullapudi et al. It consists of 30 videos, each 30 minutes in duration and at least 720p resolution. (There is 900 total minutes of video). The streams were obtained from publicly available streams on Youtube.

This work was supported by NSF IIS-1422767.