Video (language) modeling: a baseline for generative models of natural videos

TL;DR

This paper introduces a baseline for unsupervised video feature learning by adapting language modeling techniques to predict missing or future frames, capturing complex motion and deformation patterns.

Contribution

It adapts language modeling methods to the vision domain for video prediction, demonstrating effective motion prediction on natural videos.

Findings

Model predicts non-trivial motions in short video sequences

Effective learning of spatial and temporal correlations

Applicable to both filling and generation tasks

Abstract

We propose a strong baseline model for unsupervised feature learning using video data. By learning to predict missing frames or extrapolate future frames from an input video sequence, the model discovers both spatial and temporal correlations which are useful to represent complex deformations and motion patterns. The models we propose are largely borrowed from the language modeling literature, and adapted to the vision domain by quantizing the space of image patches into a large dictionary. We demonstrate the approach on both a filling and a generation task. For the first time, we show that, after training on natural videos, such a model can predict non-trivial motions over short video sequences.