CircleGAN: Generative Adversarial Learning across Spherical Circles

TL;DR

CircleGAN introduces a spherical circle-based discriminator that enhances the realism and diversity of generated samples by leveraging a structured hypersphere embedding, outperforming existing methods in both unconditional and conditional generation tasks.

Contribution

The paper proposes a novel discriminator using spherical circles on a hypersphere for improved diversity and realism in GANs, extending it to conditional settings with class-wise discrimination.

Findings

Achieves state-of-the-art results on standard benchmarks.

Improves diversity of generated samples through hypersphere embedding.

Enhances realism of generated images in both unconditional and conditional GANs.

Abstract

We present a novel discriminator for GANs that improves realness and diversity of generated samples by learning a structured hypersphere embedding space using spherical circles. The proposed discriminator learns to populate realistic samples around the longest spherical circle, i.e., a great circle, while pushing unrealistic samples toward the poles perpendicular to the great circle. Since longer circles occupy larger area on the hypersphere, they encourage more diversity in representation learning, and vice versa. Discriminating samples based on their corresponding spherical circles can thus naturally induce diversity to generated samples. We also extend the proposed method for conditional settings with class labels by creating a hypersphere for each category and performing class-wise discrimination and update. In experiments, we validate the effectiveness for both unconditional and conditional generation on standard benchmarks, achieving the state of the art.