VideoFlow: A Conditional Flow-Based Model for Stochastic Video Generation

TL;DR

VideoFlow introduces a flow-based probabilistic model for multi-frame stochastic video prediction, enabling direct likelihood optimization and high-quality uncertain future predictions.

Contribution

It is the first to apply normalizing flows to multi-frame video prediction, modeling latent space dynamics for efficient and high-quality stochastic video generation.

Findings

Flow-based models produce diverse and realistic video predictions.

The approach allows direct likelihood optimization for better training.

Results outperform some existing probabilistic video prediction methods.

Abstract

Generative models that can model and predict sequences of future events can, in principle, learn to capture complex real-world phenomena, such as physical interactions. However, a central challenge in video prediction is that the future is highly uncertain: a sequence of past observations of events can imply many possible futures. Although a number of recent works have studied probabilistic models that can represent uncertain futures, such models are either extremely expensive computationally as in the case of pixel-level autoregressive models, or do not directly optimize the likelihood of the data. To our knowledge, our work is the first to propose multi-frame video prediction with normalizing flows, which allows for direct optimization of the data likelihood, and produces high-quality stochastic predictions. We describe an approach for modeling the latent space dynamics, and demonstrate that flow-based generative models offer a viable and competitive approach to generative modelling of video.