VCT: A Video Compression Transformer

TL;DR

This paper introduces VCT, a transformer-based approach for neural video compression that simplifies architecture by modeling frame dependencies directly, outperforming previous methods on standard datasets.

Contribution

The paper presents a novel transformer-based model for video compression that eliminates complex priors and achieves superior performance.

Findings

Outperforms previous video compression methods on standard datasets

Learns to handle complex motion patterns from data

Easy to implement and open-sourced

Abstract

We show how transformers can be used to vastly simplify neural video compression. Previous methods have been relying on an increasing number of architectural biases and priors, including motion prediction and warping operations, resulting in complex models. Instead, we independently map input frames to representations and use a transformer to model their dependencies, letting it predict the distribution of future representations given the past. The resulting video compression transformer outperforms previous methods on standard video compression data sets. Experiments on synthetic data show that our model learns to handle complex motion patterns such as panning, blurring and fading purely from data. Our approach is easy to implement, and we release code to facilitate future research.