Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a 2D perceptual wrapper that enhances the texture synthesis capabilities of 3D Gaussian Splatting (3DGS) by utilizing a lightweight synthesis network conditioned on pseudo-random Gaussian noise. The approach addresses the challenge of generating high-frequency textures, particularly in memory-constrained and rate-distortion-optimized scenarios, by focusing on matching local feature statistics rather than strict pixel-wise fidelity. Experimental results show that this method significantly improves perceptual quality while reducing file and model sizes across various 3DGS implementations.
Achieving superior texture synthesis in 3D Gaussian Splatting, our method cuts file sizes while enhancing perceptual quality, even in resource-limited settings.
While 3D Gaussian Splatting (3DGS) achieves impressive real-time rendering, it frequently struggles to synthesize high-frequency textures, a limitation heavily exacerbated in memory-constrained and rate-distortion-optimized (RDO) pipelines. To address this, we propose a versatile 2D perceptual wrapper that enhances the rendered outputs of existing 3DGS representations in a content- and view-dependent manner. Our method leverages a lightweight synthesis network conditioned on pseudo-random Gaussian noise to synthesize perceptually plausible textures. Supervised by Wasserstein Distortion, the network learns to match local feature statistics rather than strictly enforcing pixel-wise reconstruction fidelity, effectively mitigating the blurriness inherent in standard frameworks. We demonstrate the broad applicability of our plug-and-play approach across vanilla, memory-constrained, and RDO 3DGS methods. Comprehensive subjective and objective experiments confirm that our method significantly improves over existing baselines, yielding superior perceptual quality at sharply reduced file or model sizes.