AnimGAN: A Spatiotemporally-Conditioned Generative Adversarial Network for Character Animation

TL;DR

This paper introduces AnimGAN, a novel spatiotemporally-conditioned GAN that generates realistic, semantically relevant character animations with smooth dynamics, controlled by user-defined behaviors, trained on a large gesture dataset.

Contribution

The paper presents a new GAN architecture with LSTM and graph ConvNet components for generating controllable, realistic character animation sequences.

Findings

Produces plausible, realistic animations

Maintains semantic relevance and spatiotemporal smoothness

Outperforms traditional conditional GANs in quality

Abstract

Producing realistic character animations is one of the essential tasks in human-AI interactions. Considered as a sequence of poses of a humanoid, the task can be considered as a sequence generation problem with spatiotemporal smoothness and realism constraints. Additionally, we wish to control the behavior of AI agents by giving them what to do and, more specifically, how to do it. We proposed a spatiotemporally-conditioned GAN that generates a sequence that is similar to a given sequence in terms of semantics and spatiotemporal dynamics. Using LSTM-based generator and graph ConvNet discriminator, this system is trained end-to-end on a large gathered dataset of gestures, expressions, and actions. Experiments showed that compared to traditional conditional GAN, our method creates plausible, realistic, and semantically relevant humanoid animation sequences that match user expectations.