AdaptiveSplat:Texture Aware Controllable 3D Gaussian Allocation for Feed-Forward Reconstruction

2026-07-05

AI Summary

This work addresses the inefficiency of existing feed-forward 3D Gaussian reconstruction methods, which often suffer from redundant primitives, while direct pruning tends to introduce artifacts and compromise the feed-forward property. To overcome these limitations, the paper introduces, for the first time, a texture-aware mechanism that explicitly controls the number of Gaussian primitives based on local texture cues. The proposed unified framework integrates a texture estimation module, a texture-aware pruning strategy, and an adaptive Gaussian attribute prediction head, enabling controllable sparsification without fine-tuning while preserving purely feed-forward inference. Extensive experiments demonstrate significant improvements in both reconstruction quality and efficiency across multiple benchmarks—including RE10K, ACID, DL3DV, Tanks and Temples, and DTU—and ablation studies confirm the effectiveness of each component.

Abstract

Current feed-forward 3D reconstruction methods predict pixel aligned Gaussian primitives, resulting in highly redundant representations. A natural solution is to prune the redundant Gaussians, but naive pruning introduces severe artifacts and often requires inference time fine-tuning, breaking the feed-forward paradigm. Based on previous works, high frequency regions require more Gaussian primitives, while low frequency regions can be represented with significantly fewer primitives. Motivated by this, we propose a novel approach to explicitly control the number of Gaussians by leveraging local texture information. Our approach achieves this through three key components: (1) texture estimation to capture spatial variation in scene detail, (2) texture-aware pruning that removes redundant Gaussians from low frequency regions, and (3) an adaptive Gaussian head that predicts the modified attributes of the retained primitives without breaking the feed-forward paradigm. Experiments on RE10K, ACID, DL3DV, Tanks and Temples, and DTU demonstrate the effectiveness of our approach, while ablation studies validate the contributions of its key components.

Problem

• 3D reconstruction
• Gaussian primitives
• redundancy
• feed-forward paradigm
• texture-aware pruning

Innovation

• texture-aware pruning
• adaptive Gaussian allocation
• feed-forward 3D reconstruction
• Gaussian splatting
• spatial detail estimation