Geometric Autoencoder for Diffusion Models
This work addresses a key bottleneck in latent diffusion models for high-resolution visual generation, offering a principled framework that improves generative efficiency and quality.
The paper tackles the challenge of unifying semantic discriminability, reconstruction fidelity, and latent compactness in latent diffusion models by proposing the Geometric Autoencoder (GAE), which achieves a gFID of 1.82 at 80 epochs and 1.31 at 800 epochs on ImageNet-1K 256x256, surpassing existing state-of-the-art methods.
Latent diffusion models have established a new state-of-the-art in high-resolution visual generation. Integrating Vision Foundation Model priors improves generative efficiency, yet existing latent designs remain largely heuristic. These approaches often struggle to unify semantic discriminability, reconstruction fidelity, and latent compactness. In this paper, we propose Geometric Autoencoder (GAE), a principled framework that systematically addresses these challenges. By analyzing various alignment paradigms, GAE constructs an optimized low-dimensional semantic supervision target from VFMs to provide guidance for the autoencoder. Furthermore, we leverage latent normalization that replaces the restrictive KL-divergence of standard VAEs, enabling a more stable latent manifold specifically optimized for diffusion learning. To ensure robust reconstruction under high-intensity noise, GAE incorporates a dynamic noise sampling mechanism. Empirically, GAE achieves compelling performance on the ImageNet-1K $256 \times 256$ benchmark, reaching a gFID of 1.82 at only 80 epochs and 1.31 at 800 epochs without Classifier-Free Guidance, significantly surpassing existing state-of-the-art methods. Beyond generative quality, GAE establishes a superior equilibrium between compression, semantic depth and robust reconstruction stability. These results validate our design considerations, offering a promising paradigm for latent diffusion modeling. Code and models are publicly available at https://github.com/sii-research/GAE.