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

About

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.

Aleksander Ogonowski, Konrad Klimaszewski, Przemys\law Rokita · 2026

Related benchmarks

Task	Dataset	Result
Image Generation	LSUN Church 256x256	FID12.11	10
Class-conditional image synthesis	LSUN Rooms 128x128 (Global)	FID17.79	6
Class-conditional image synthesis	LSUN Rooms 128x128 Bedroom	FID21.39	6
Class-conditional image synthesis	LSUN Rooms Kitchen 128x128	FID23.32	6
Image Synthesis	AFHQ 512x512 Global	FID8.81	6
Image Synthesis	AFHQ Global 256x256	FID10.29	5
Image Synthesis	AFHQ 256x256 Cat	FID8.87	5
Image Synthesis	AFHQ 256x256 Dog	FID26.51	5
Image Synthesis	AFHQ Wild 256x256	FID6.07	5
Image Generation	AFHQ Cats 512x512	FID (512x512)12.63	5