Hybrid Latents -- Geometry-Appearance-Aware Surfel Splatting

2026-04-16

AI Summary

This work addresses the reconstruction inaccuracies and rendering inefficiencies in NeRF-like models caused by the tight coupling between geometry and appearance. To this end, we propose a hybrid Gaussian–hash-grid radiance representation that explicitly decouples low-frequency geometry from high-frequency texture through frequency-aware embedding. Each Gaussian is augmented with a latent feature vector, jointly optimized alongside hash-grid features. The method further incorporates hard opacity decay, a binary cross-entropy opacity loss, and a probabilistic pruning strategy to automatically sparsify and refine the Gaussian primitives. Experiments demonstrate that our approach achieves superior reconstruction fidelity and rendering efficiency on both synthetic and real-world datasets, using an order of magnitude fewer Gaussians than existing methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Mixed Discrete/Continuous Search
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Security and Privacy: Data transparency and provenance

Abstract

We introduce a hybrid Gaussian-hash-grid radiance representation for reconstructing 2D Gaussian scene models from multi-view images. Similar to NeST splatting, our approach reduces the entanglement between geometry and appearance common in NeRF-based models, but adds per-Gaussian latent features alongside hash-grid features to bias the optimizer toward a separation of low- and high-frequency scene components. This explicit frequency-based decomposition reduces the tendency of high-frequency texture to compensate for geometric errors. Encouraging Gaussians with hard opacity falloffs further strengthens the separation between geometry and appearance, improving both geometry reconstruction and rendering efficiency. Finally, probabilistic pruning combined with a sparsity-inducing BCE opacity loss allows redundant Gaussians to be turned off, yielding a minimal set of Gaussians sufficient to represent the scene. Using both synthetic and real-world datasets, we compare against the state of the art in Gaussian-based novel-view synthesis and demonstrate superior reconstruction fidelity with an order of magnitude fewer primitives.

Problem

• geometry-appearance entanglement
• high-frequency texture compensation
• geometric error
• scene reconstruction
• novel-view synthesis

Innovation

• hybrid latent representation
• geometry-appearance disentanglement
• frequency-based decomposition
• probabilistic pruning
• Gaussian splatting