DSS-GAN: Directional State Space GAN with Mamba backbone for Class-Conditional Image Synthesis

TL;DR

DSS-GAN introduces a hierarchical generator backbone with Directional Latent Routing, enabling more structured and class-aware image synthesis, outperforming existing models on key quality metrics.

Contribution

The paper proposes DSS-GAN with a novel Directional Latent Routing mechanism and Mamba backbone, advancing class-conditional image synthesis techniques.

Findings

Achieves better FID, KID, and precision-recall scores than StyleGAN2-ADA.

Latent space analysis shows directional subvectors encode structured variations.

Directional conditioning couples class and latent information along spatial axes.

Abstract

We present DSS-GAN, the first generative adversarial network to employ Mamba as a hierarchical generator backbone for noise-to-image synthesis. The central contribution is Directional Latent Routing (DLR), a novel conditioning mechanism that decomposes the latent vector into direction-specific subvectors, each jointly projected with a class embedding to produce a feature-wise affine modulation of the corresponding Mamba scan. Unlike conventional class conditioning that injects a global signal, DLR couples class identity and latent structure along distinct spatial axes of the feature map, applied consistently across all generative scales. DSS-GAN achieves improved FID, KID, and precision-recall scores compared to StyleGAN2-ADA across multiple tested datasets. Analysis of the latent space reveals that directional subvectors exhibit measurable specialization: perturbations along individual components produce structured, direction-correlated changes in the synthesized image.