Autoregressive GAN for Semantic Unconditional Head Motion Generation

TL;DR

This paper introduces an autoregressive GAN model for generating realistic, long-duration head motion sequences from a single pose, improving motion quality and stability over previous methods.

Contribution

The paper proposes a novel autoregressive GAN architecture with multi-scale discrimination for unconditional head motion generation from semantic inputs.

Findings

Outperforms state-of-the-art models in head motion realism

Produces smooth, long-duration motion sequences

Reduces mode collapse and error accumulation

Abstract

In this work, we address the task of unconditional head motion generation to animate still human faces in a low-dimensional semantic space from a single reference pose. Different from traditional audio-conditioned talking head generation that seldom puts emphasis on realistic head motions, we devise a GAN-based architecture that learns to synthesize rich head motion sequences over long duration while maintaining low error accumulation levels.In particular, the autoregressive generation of incremental outputs ensures smooth trajectories, while a multi-scale discriminator on input pairs drives generation toward better handling of high- and low-frequency signals and less mode collapse.We experimentally demonstrate the relevance of the proposed method and show its superiority compared to models that attained state-of-the-art performances on similar tasks.