StyleVideoGAN: A Temporal Generative Model using a Pretrained StyleGAN

TL;DR

StyleVideoGAN introduces a novel video synthesis method that leverages a pretrained StyleGAN for high-quality frame generation and a separate temporal model trained on latent codes, drastically reducing training data and resources needed.

Contribution

The paper presents a new approach combining pretrained StyleGAN with a temporal model trained on latent codes, enabling efficient and high-quality video synthesis with minimal training data.

Findings

Achieves high-quality video generation with only 10 minutes of training data.

Can generate videos of new subjects not seen during training.

Reduces training time to approximately 6 hours.

Abstract

Generative adversarial models (GANs) continue to produce advances in terms of the visual quality of still images, as well as the learning of temporal correlations. However, few works manage to combine these two interesting capabilities for the synthesis of video content: Most methods require an extensive training dataset to learn temporal correlations, while being rather limited in the resolution and visual quality of their output. We present a novel approach to the video synthesis problem that helps to greatly improve visual quality and drastically reduce the amount of training data and resources necessary for generating videos. Our formulation separates the spatial domain, in which individual frames are synthesized, from the temporal domain, in which motion is generated. For the spatial domain we use a pre-trained StyleGAN network, the latent space of which allows control over the appearance of the objects it was trained for. The expressive power of this model allows us to embed our training videos in the StyleGAN latent space. Our temporal architecture is then trained not on sequences of RGB frames, but on sequences of StyleGAN latent codes. The advantageous properties of the StyleGAN space simplify the discovery of temporal correlations. We demonstrate that it suffices to train our temporal architecture on only 10 minutes of footage of 1 subject for about 6 hours. After training, our model can not only generate new portrait videos for the training subject, but also for any random subject which can be embedded in the StyleGAN space.