Video Frame Synthesis using Deep Voxel Flow

TL;DR

This paper introduces Deep Voxel Flow, a neural network approach for synthesizing new video frames through pixel flow, improving upon existing methods in quality and efficiency without requiring human supervision.

Contribution

The paper presents a novel deep network that combines optical flow and pixel hallucination for video frame synthesis, trained without supervision on any video data.

Findings

Outperforms state-of-the-art in quality and speed

Works at any video resolution

Requires no human supervision

Abstract

We address the problem of synthesizing new video frames in an existing video, either in-between existing frames (interpolation), or subsequent to them (extrapolation). This problem is challenging because video appearance and motion can be highly complex. Traditional optical-flow-based solutions often fail where flow estimation is challenging, while newer neural-network-based methods that hallucinate pixel values directly often produce blurry results. We combine the advantages of these two methods by training a deep network that learns to synthesize video frames by flowing pixel values from existing ones, which we call deep voxel flow. Our method requires no human supervision, and any video can be used as training data by dropping, and then learning to predict, existing frames. The technique is efficient, and can be applied at any video resolution. We demonstrate that our method produces results that both quantitatively and qualitatively improve upon the state-of-the-art.