Visual Dynamics: Stochastic Future Generation via Layered Cross Convolutional Networks

TL;DR

This paper introduces a probabilistic approach for future frame synthesis from a single image using a novel Cross Convolutional Network, enabling diverse and realistic motion predictions for both synthetic and real-world data.

Contribution

It presents a new probabilistic model and a Cross Convolutional Network architecture for generating multiple plausible future frames from a single image.

Findings

Model performs well on synthetic and real-world data

Network learns compact encoding of object appearance and motion

Applications include visual analogy-making and video extrapolation

Abstract

We study the problem of synthesizing a number of likely future frames from a single input image. In contrast to traditional methods that have tackled this problem in a deterministic or non-parametric way, we propose to model future frames in a probabilistic manner. Our probabilistic model makes it possible for us to sample and synthesize many possible future frames from a single input image. To synthesize realistic movement of objects, we propose a novel network structure, namely a Cross Convolutional Network; this network encodes image and motion information as feature maps and convolutional kernels, respectively. In experiments, our model performs well on synthetic data, such as 2D shapes and animated game sprites, and on real-world video frames. We present analyses of the learned network representations, showing it is implicitly learning a compact encoding of object appearance and motion. We also demonstrate a few of its applications, including visual analogy-making and video extrapolation.