CLEAR: Conflict-aware Learning via Evidence-guided Adaptive Routing for Unified Sparse-View 3D Gaussian Super-Resolution

2026-08-03

AI Summary

This work addresses the challenge of super-resolution reconstruction for 3D Gaussian splatting under sparse-view settings, where limited input data hinders accurate recovery of geometry and high-frequency details, and existing two-stage approaches suffer from error accumulation. To overcome these limitations, we propose CLEAR—the first unified single-stage framework tailored to this task—that jointly optimizes low-resolution observations with external high-resolution priors. CLEAR introduces several key innovations, including Gaussian-level conflict-aware optimization, an evidence-guided patch-to-Gaussian high-frequency routing mechanism, shared Gaussian dropout, and mid-training anchor decoupling, which collectively mitigate gradient conflicts and adaptively enhance detail restoration. Extensive experiments on both synthetic and real-world benchmarks demonstrate that CLEAR achieves significant improvements over state-of-the-art methods in 4× super-resolution, consistently advancing both rendering quality and geometric fidelity.

Abstract

Sparse-view 3D Gaussian Splatting Super-resolution is highly challenging since the sparse and low-resolution (LR) inputs lack sufficient geometric and high-frequency information for accurate reconstruction. To achieve high-quality reconstruction, existing sparse-view super-resolution methods adhere to two-stage pipeline that performs LR Gaussian reconstruction and then high-resolution (HR) Gaussian refinement, which directly results in stage-wise Gaussian transfer and reconstruction error accumulation. To this end, we propose CLEAR, a Conflict-aware Learning via Evidence-guided Adaptive Routing, as the first unified single-stage framework for Sparse-view 3D Gaussian Splatting Super-resolution. Specifically, CLEAR performs joint the optimization of authentic LR observations and external HR priors within a unified Gaussian representation. To mitigate the gradient conflicts introduced by sparse supervision during training, we propose a Gaussian-wise conflict-aware optimization strategy that regards the LR gradient as a reliable anchor and applies evidence-conditioned soft correction only to severe HR conflicts. Moreover, to recover high-frequency details, we introduce an evidence-guided Patch-to-Gaussian routing mechanism which estimates patch reliability and detail demand, lifts them into Gaussian space, and selectively routes high-frequency gradients and densification. Finally, we employ shared Gaussian dropout and a detached mid-training anchoring to enhance the robustness of training framework. Extensive experiments on both synthetic and real-world $4\times$ super-resolution benchmarks demonstrate that CLEAR consistently achieves state-of-the-art rendering quality and superior geometric fidelity.

Problem

• Sparse-view
• 3D Gaussian Splatting
• Super-resolution
• High-frequency Detail Recovery
• Reconstruction Error Accumulation

Innovation

• unified single-stage framework
• conflict-aware optimization
• evidence-guided routing
• Gaussian splatting super-resolution
• gradient conflict mitigation