DenoiseSplat: Feed-Forward Gaussian Splatting for Noisy 3D Scene Reconstruction

TL;DR

DenoiseSplat introduces a feed-forward Gaussian splatting approach that effectively reconstructs 3D scenes from noisy multi-view images, outperforming existing methods without requiring 3D ground truth.

Contribution

The paper presents a novel end-to-end trainable 3D Gaussian splatting method that handles noisy inputs using only clean 2D supervision, with a new noisy-clean benchmark for evaluation.

Findings

Outperforms vanilla MVSplat and two-stage baselines in PSNR/SSIM and LPIPS.

Effective across various noise types and levels.

Built a large-scale noisy-clean benchmark on RE10K.

Abstract

3D scene reconstruction and novel-view synthesis are fundamental for VR, robotics, and content creation. However, most NeRF and 3D Gaussian Splatting pipelines assume clean inputs and degrade under real noise and artifacts. We therefore propose DenoiseSplat, a feed-forward 3D Gaussian splatting method for noisy multi-view images. We build a large-scale, scene-consistent noisy--clean benchmark on RE10K by injecting Gaussian, Poisson, speckle, and salt-and-pepper noise with controlled intensities. With a lightweight MVSplat-style feed-forward backbone, we train end-to-end using only clean 2D renderings as supervision and no 3D ground truth. On noisy RE10K, DenoiseSplat outperforms vanilla MVSplat and a strong two-stage baseline (IDF + MVSplat) in PSNR/SSIM and LPIPS across noise types and levels.