Search papers, labs, and topics across Lattice.
This paper introduces PixGS, a novel single-stage pipeline for generating high-quality 3D Gaussian Splats (3DGS) that addresses the limitations of existing multi-stage diffusion models. By leveraging pixel-space diffusion, PixGS eliminates the need for lossy latent compression and directly denoises 3D Gaussian attributes, resulting in improved appearance and geometry regularization. Experimental results show that PixGS not only surpasses state-of-the-art methods in quality but also achieves rapid inference speeds, making it a practical solution for 3D content generation.
Bypassing lossy latent compression, PixGS generates high-quality 3D Gaussian Splats in a single stage, outperforming multi-stage pipelines in both quality and speed.
Recent advances in 3D content generation from text or images have achieved impressive results, yet view inconsistency from 2D generators and the scarcity of high-quality 3D data remain significant bottlenecks. Existing solutions typically adapt large-scale pre-trained text-to-image latent diffusion models to generate 3D Gaussian Splats (3DGS). However, these approaches often rely on training complex cascade pipelines that are computationally expensive and scalability-limited. Most critically, the quality of generated 3D assets is inherently constrained by each component capacity and compressed latent space, leading to decoding artifacts and accumulated errors. To address these limitations, we propose PixGS, a single-stage pipeline for direct high-quality 3DGS generation, which leverages recent advances in pixel-space diffusion to bypass lossy latent compression while still benefiting from the vast 2D generative priors. By directly denoising 3D Gaussian attributes at each timestep, our method enables precise, splat-level regularization of both appearance and geometry. Furthermore, we introduce a comprehensive supervision strategy that incorporates surface normals, depth, and high-frequency structural information, which is often overlooked in prior works. Experiments demonstrate that PixGS outperforms current state-of-the-art methods while maintaining a fast inference speed (1s on a single A100 GPU), offering a robust and efficient alternative to multi-stage generation pipelines.
