This page is an AI translation of the Japanese original. If anything differs, the Japanese version prevails.
SemLite-GS: A 3DGS Method That Turns Only the Buildings in UAV Oblique Images into Lightweight Meshes
Tengfei Wang and colleagues have published SemLite-GS on arXiv, a method that trains 3DGS in parallel for each building from UAV oblique images to produce lightweight meshes. The code is also public, and the authors say it can shorten building reconstruction time compared with whole-scene approaches.
This article was generated by AI and published automatically without prior human review.
- Announced
- 2026.11.15 (local time)
- Use cases
- Infrastructure, Research, Production
- Sources
- 1 (en) (1 primary)
Key points
- A framework that updates only the target Gaussians for each building and trains multiple buildings in parallel
- Building IDs are unified using an orthographic reference map and mapped to oblique images to create regions of interest
- Source code is public on GitHub. Results such as the time reduction are values announced in the paper
Contents
What happened
The paper "SemLite-GS: Semantic-Selective Parallel Gaussian Splatting for Lightweight Building Reconstruction from UAV Oblique Imagery" was submitted to arXiv on November 15, 2026. The authors are Tengfei Wang, Xin Wang, Zihan Li, Quanjian Ji, Yifei Yu, Dong Lin and Zongqian Zhan.
How it works
The authors position conventional lightweight building modeling as often requiring several stages: dense reconstruction of the whole scene, building extraction, and mesh simplification. SemLite-GS replaces this pipeline with a building-centered framework.
By the numbers
| Input | Large-scale UAV oblique images |
|---|---|
| Output | Lightweight building meshes |
| Training method | Building-selective parallel PGSR |
| Meshing | SDF optimization (guided by planar information) |
| Code | GitHub (IRIS-LAB-whu/SemLite-GS) |
Why it matters
What can be confirmed from the source material is limited to the claims in the paper. The work aims to efficiently create lightweight building models for urban spatial analysis from UAV oblique imagery, and it is likely to draw interest in urban modeling, surveying and digital twins around infrastructure.
Caveats and open questions
- Specific figures on the reduction in reconstruction time or on shape preservation are not given in the abstract in the source material.
- The names of the datasets used for evaluation and details of the compared methods cannot be confirmed from the source material.
- The paper is a preprint on arXiv, and its peer-review status cannot be confirmed from the source material.
- The code's operating environment and license have not been checked.
Glossary
- UAV oblique images
- Images captured from an oblique angle by drones or similar platforms. They can also capture the sides of buildings.
- PGSR
- A 3DGS-based method that is plane-aware and improves geometric accuracy.
- SDF (signed distance field)
- A representation that expresses the signed distance from each point in space to a surface. It is used for mesh extraction.
- Mesh simplification (decimation)
- A process that reduces the polygon count while preserving the shape.
Sources
Primary sources (publisher, paper, repository)
- SemLite-GS: Semantic-Selective Parallel Gaussian Splatting for Lightweight Building Reconstruction from UAV Oblique ImageryarXiv(arxiv.org)・en
The X post that surfaced this story
This article was generated in Japanese by AI from the sources listed (primarily primary sources) and published automatically, without prior human review, after automated checks such as source matching (how we use AI). Figures and specifications are as stated by the publishers and have not been independently verified by us. Please consult the original sources. To report an error, use our contact form; we will review it promptly and correct it. This page is an AI translation of the Japanese original. If anything differs, the Japanese version prevails. Japanese original
From the operatorガウシアンスプラッティングとは(IZUTSUYA Inc.)