Mobile system creates compact 3D Gaussian reconstructions efficiently
PocketSplat: Mobile Gaussian Reconstruction via World-Space Latent Allocatio
Computer Vision and Pattern Recognition
Summary
Creating detailed 3D shapes on mobile devices is hard because they have limited memory and computing power. The authors present PocketSplat, a method that makes small, efficient 3D models called Gaussians directly on phones. This method smartly picks and processes only the important parts of the data to keep quality high without using too much memory. They show PocketSplat works well on various tests, runs fast on an iPhone, and makes better, smaller 3D models than some existing methods.
What this means in practice
- •For mobile app developers: Build apps that create and use efficient 3D Gaussian models directly on phones without exceeding memory limits.
- •For augmented reality teams: Integrate compact 3D Gaussian assets reconstructed on-device to improve AR experiences with faster processing and high-quality visuals.
Authors
Wenzhi Guo, Xianda Chen, Dongxuan Chen, Guangchi Fang, Bing Wang
Abstract
Mobile Gaussian reconstruction must satisfy two requirements: the reconstruction model must execute within a device resource envelope, and the resulting Gaussian asset must expose a representation size suited to downstream mobile use. Existing feed-forward Gaussian reconstructors commonly decode dense, image-aligned candidates whose final cardinality is implicitly determined by the input resolution and number of views. We present PocketSplat, a feed-forward framework for budgeted mobile Gaussian asset construction. Given a prescribed output budget, PocketSplat organizes dense geometry-aware latent candidates in predicted world space, allocates exact integer capacity across local latent cells, and decodes complete Gaussian attributes only for retained candidates. Cell-conditioned latent fusion aggregates repeated multi-view evidence before decoding, while spatial responsibility decoding adapts Gaussian support after local sparsification. Experiments on DL3DV and out-of-distribution benchmarks establish a strong quality--budget trade-off against feed-forward Gaussian reconstruction baselines. On Mip-NeRF 360, PocketSplat executes directly on a target iPhone and constructs compact, higher-quality Gaussian assets substantially faster than a deployable streamed MVSplat variant; native MVSplat and DepthSplat exceed the device memory budget.