AI Paper Atlas

Research map / 3D vision / Gaussian splatting

MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images

Yuedong Chen, Haofei Xu, Chuanxia Zheng, Bohan Zhuang, Marc Pollefeys, Andreas Geiger, Tat-Jen Cham, Jianfei Cai

ECCV 2024 · 3D vision › Gaussian splatting · approach: Compact and efficient Gaussian splatting · 409 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add cost volume (plane-sweep feature matching) for depth (ablation)MVSplat base · RealEstate10KPSNR: 22.83 -> 26.12Improved
add cross-view attention in the Transformer (ablation)MVSplat base · RealEstate10KPSNR: 25.19 -> 26.12Improved
add cost volume refinement U-Net (ablation)MVSplat base · RealEstate10KPSNR: 25.45 -> 26.12Improved
add depth refinement U-Net (ablation)MVSplat base · RealEstate10KPSNR: 26.12 -> 26.39Improved
replace pixelSplat depth probability branch with MVSplat cost volume depth (ablation)pixelSplat · RealEstate10KPSNR: 25.89 -> 26.63Improved
use Swin Transformer instead of Epipolar Transformer (ablation)MVSplat · RealEstate10KPSNR: 26.09 -> 26.12 PSNR; time 0.055 -> 0.038 sNo clear effect
predict 3 Gaussians per pixel instead of 1 (ablation)MVSplat · RealEstate10KPSNR: 26.39 -> 26.54 PSNR; render 0.0015 -> 0.0023 sMixed
random init instead of UniMatch pretrained init (300K) (ablation)MVSplat · RealEstate10KPSNR: 26.39 -> 26.01Hurt
train MVSplat instead of pixelSplat for zero-shot transfer to DTURealEstate10K-trained models · DTU (zero-shot)LPIPS: 0.560 -> 0.385Improved
increase input views from 2 to 3 at test timepixelSplat · DTU (zero-shot)PSNR: 12.89 -> 12.52Hurt
increase input views from 2 to 3 at test timeMVSplat · DTU (zero-shot)PSNR: 13.94 -> 14.30Improved

Later papers that compared with this method

65 later accepted papers report MVSplat as a baseline in their results tables; 62 of them report a better number on most of the results they share with it.

57 more not listed.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy