Prediction Under Uncertainty with Error-Encoding Networks
This page includes video generation examples using the EEN model described in this paper.
For each sequence of frames, the first 4 frames are given and the last 4 frames are generated.
Different sequences are generated for the same set of initial frames by conditioning on different latent variables.
These latent variables are learned in an unsupervised manner from the videos themselves, no external information is used.
Code and trained models used to generate these videos are available here.