DSS-GAN: Directional State Space GAN with Mamba backbone for Class-Conditional Image Synthesis
This work addresses image generation for AI applications by improving conditioning in GANs, though it appears incremental as it builds on existing GAN frameworks with a new method.
The paper tackles class-conditional image synthesis by introducing DSS-GAN, which uses a Mamba backbone and a novel conditioning mechanism called Directional Latent Routing (DLR) to decompose latent vectors into direction-specific subvectors, achieving improved FID, KID, and precision-recall scores compared to StyleGAN2-ADA on multiple datasets.
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.