Problem
Research questions and friction points this paper is trying to address.
Anti-Aliasing
Feed-forward 3D Gaussian Splatting
Rendering Artifacts
Novel View Synthesis
Sparse-view 3D Reconstruction
AA-Splat: Anti-Aliased Feed-forward Gaussian Splatting
🤖 AI Summary
Existing feed-forward 3D Gaussian splatting methods suffer from severe aliasing at non-training resolutions due to the use of inappropriate screen-space expansion filters. To address this issue, this work proposes AA-Splat, which introduces the multi-view maximum frequency bound into feed-forward 3D Gaussian reconstruction for the first time and designs an Opacity-Balanced Band-Limited (OBBL) mechanism to enable anti-aliased rendering at arbitrary resolutions. By integrating 3D band-limited post-filtering with opacity balancing, the method effectively eliminates degenerate Gaussians and mitigates pixel-aligned Gaussian overlap artifacts. Experiments demonstrate that AA-Splat achieves an average PSNR improvement of 5.4–7.5 dB over the state-of-the-art DepthSplat across a resolution range spanning from 4× down to 1/4× of the original training resolution in novel view synthesis.
📝 Abstract
Feed-forward 3D Gaussian Splatting (FF-3DGS) emerges as a fast and robust solution for sparse-view 3D reconstruction and novel view synthesis (NVS). However, existing FF-3DGS methods are built on incorrect screen-space dilation filters, causing severe rendering artifacts when rendering at out-of-distribution sampling rates. We firstly propose an FF-3DGS model, called AA-Splat, to enable robust anti-aliased rendering at any resolution. AA-Splat utilizes an opacity-balanced band-limiting (OBBL) design, which combines two components: a 3D band-limiting post-filter integrates multi-view maximal frequency bounds into the feed-forward reconstruction pipeline, effectively band-limiting the resulting 3D scene representations and eliminating degenerate Gaussians; an Opacity Balancing (OB) to seamlessly integrate all pixel-aligned Gaussian primitives into the rendering process, compensating for the increased overlap between expanded Gaussian primitives. AA-Splat demonstrates drastic improvements with average 5.4$\sim$7.5dB PSNR gains on NVS performance over a state-of-the-art (SOTA) baseline, DepthSplat, at all resolutions, between $4\times$ and $1/4\times$. Code will be made available.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Anti-Aliasing
Feed-forward Gaussian Splatting
Band-Limiting
Opacity Balancing
Novel View Synthesis
🔎 Similar Papers
No similar papers found.
