I present the paper “Learning Temporal Transformations From Time-Lapse Videos“, which is written by Yipin Zhou and Tamara L. Berg and published in ECCV 2016, to the members of Enhanced Visual Experience (EVE) team. In this paper, they design a deep network that learn about four different natural transformations: melting, baking, blooming, and rooting. Here you can find the link to the project page. They also published their dataset which is available in the project page.
You also can find my presentation here.