Awesome 3DGS for GeoAI

Bilingual resource list for 3D Gaussian Splatting, spatial AI, disaster mapping, urban reconstruction, and field-ready 3D scene capture.

面向 3D Gaussian Splatting、空间智能、灾害制图、城市重建与现场三维场景采集的中英文资源库。

Why This Matters

3D Gaussian Splatting (3DGS) can reconstruct photorealistic 3D scenes from 2D images and render them in real time. For GeoAI and disaster research, this is useful because field photos, drone images, and street-level videos can become inspectable 3D evidence rather than isolated 2D frames.

For disaster mapping, a practical question is: can a fast 3DGS pipeline help analysts inspect damaged buildings, blocked roads, debris, flood marks, and scene context after an event? The near-term research direction is to compare Gaussian splat scenes with orthomosaics, LiDAR, segmentation maps, and damage labels.

3D Gaussian Splatting 可以从二维图像重建逼真的三维场景，并支持实时渲染。对于 GeoAI 和灾害研究，它的价值在于：现场照片、无人机影像、街景视频不再只是分散的二维证据，而可以转化为可浏览、可检查、可标注的三维场景。

对灾害制图来说，一个直接的问题是：能否用快速 3DGS 流程帮助研究人员检查建筑损伤、道路阻断、碎片堆积、洪水痕迹和灾后环境上下文？近期可以做的研究方向，是将 Gaussian splat 场景与正射影像、LiDAR、分割结果和灾损标签进行对比。

Quick Start Path

1. Capture images or video with a phone, drone, or street-view platform.

2. Estimate camera poses with COLMAP, OpenSfM, RealityCapture, or an integrated mobile app.

3. Train a Gaussian splat with Nerfstudio Splatfacto, gsplat, OpenSplat, Postshot, Polycam, Luma AI, or Scaniverse.

4. Clean, crop, inspect, and publish the result with SuperSplat or a WebGL viewer.

5. Compare the reconstructed scene with geospatial products such as orthomosaics, DEMs, LiDAR point clouds, damage labels, or building footprints.

6. 使用手机、无人机或街景平台采集照片或视频。

7. 使用 COLMAP、OpenSfM、RealityCapture 或移动端集成工具估计相机位姿。

8. 使用 Nerfstudio Splatfacto、gsplat、OpenSplat、Postshot、Polycam、Luma AI 或 Scaniverse 训练 Gaussian splat。

9. 使用 SuperSplat 或 WebGL viewer 清理、裁剪、检查和发布结果。

10. 将三维场景与正射影像、DEM、LiDAR 点云、灾损标签或建筑轮廓进行对比。

Core 3DGS Implementations

Resource	Type	Why It Is Useful	中文说明
3D Gaussian Splatting for Real-Time Radiance Field Rendering	Reference implementation	Original Inria implementation for the 3DGS method.	3DGS 原始参考实现，适合理解方法源头和复现实验。
Nerfstudio	Research framework	Modular framework for NeRF and Gaussian Splatting workflows, including Splatfacto.	研究型框架，适合做可控实验、替换模块和对比方法。
Nerfstudio Splatfacto	3DGS method	Nerfstudio's practical Gaussian Splatting pipeline.	Nerfstudio 中更容易上手的 Gaussian Splatting 方法。
gsplat	Library	CUDA-accelerated Gaussian Splatting library for research and scalable training.	面向研究和规模化训练的高性能 3DGS 库。
OpenSplat	Open-source implementation	C++ implementation that supports COLMAP, OpenSfM, ODM, and Nerfstudio-style inputs.	C++ 开源实现，适合和摄影测量、无人机重建流程连接。
NVIDIA Instant-NGP	Adjacent baseline	Fast neural graphics primitives and NeRF baseline for comparison.	不是 3DGS，但适合作为快速神经渲染和 NeRF 对照基线。

Capture, Scan, and Production Tools

Resource	Type	Why It Is Useful	中文说明
Niantic Spatial	Spatial AI company	Builds real-world spatial intelligence, mapping, VPS, and mobile 3D capture tools.	Niantic 的空间智能方向，和现实世界三维地图、VPS、移动端重建紧密相关。
Scaniverse	Mobile scanning app	Phone-based 3D capture app that supports Gaussian splats and mesh workflows.	Niantic 的手机扫描工具，适合快速采集现场三维场景。
Niantic Large Geospatial Model	Spatial AI research	Direction for learning geospatial representations from large-scale real-world scans.	面向大规模现实世界扫描数据的空间表征学习方向。
Niantic Visual Positioning System	Localization platform	Useful context for aligning visual scenes with real-world locations.	用于将视觉场景与真实地理位置对齐的 VPS 平台。
Luma AI Interactive Scenes	Web and capture workflow	Creator-friendly path for capturing and sharing interactive 3D scenes.	面向创作者的交互式三维场景采集与分享流程。
Luma WebGL Library	Viewer library	JavaScript library for embedding Luma captures on the web.	可将 Luma 三维内容嵌入网页的前端库。
Polycam Gaussian Splats	Web capture workflow	Upload images or videos and create Gaussian splats through a hosted workflow.	通过网页上传图像或视频生成 Gaussian splats，适合快速试验。
Jawset Postshot	Desktop production tool	Desktop tool for training and editing radiance fields and Gaussian splats.	桌面端 3DGS/辐射场制作工具，适合可视化生产流程。
PlayCanvas SuperSplat	Editor and viewer	Browser-based editor for cleaning, cropping, optimizing, and publishing splats.	浏览器端 splat 编辑器，适合清理、裁剪、压缩和展示。
SuperSplat GitHub	Open-source editor	Source code for the PlayCanvas SuperSplat editor.	SuperSplat 的开源代码，适合部署和二次开发。
Epic RealityScan	Mobile scanning app	Mobile photogrammetry and 3D capture app from Epic Games.	Epic 的移动端摄影测量与三维扫描工具。
RealityCapture	Photogrammetry software	Production photogrammetry software for high-quality 3D reconstruction.	高质量摄影测量软件，可作为 3DGS 上游重建工具。
Agisoft Metashape	Photogrammetry software	Mature commercial tool for SfM, dense reconstruction, and geospatial workflows.	成熟商业摄影测量工具，常用于无人机和测绘场景。
COLMAP	SfM and MVS	Standard open-source tool for camera pose estimation and sparse or dense reconstruction.	常用开源 SfM/MVS 工具，是许多 3DGS 流程的上游。

Large-Scale, Urban, and Geospatial 3DGS

Resource	Type	Why It Is Useful	中文说明
CityGaussian	Research code	Explores large-scale scene reconstruction with Gaussian Splatting.	面向城市或大场景的 Gaussian Splatting 研究代码。
Street Gaussians	Research code	Dynamic urban street-scene reconstruction with Gaussian primitives.	面向城市街景、自动驾驶场景和动态对象的 3DGS 方向。
Horizon-GS	Research project	Targets large-scale urban scene reconstruction.	面向大规模城市重建的 3DGS 项目。
Mega-NeRF	Large-scale NeRF baseline	Useful comparison point for city-scale neural rendering before 3DGS.	城市尺度 NeRF 代表性基线，可用于和 3DGS 路线比较。
Urban Radiance Fields	Urban NeRF baseline	Relevant prior work for reconstructing large outdoor scenes.	大型户外城市重建的相关前序工作。

Industry Deployments and Real-World Adoption

Resource	Type	Why It Is Useful	中文说明
Apartments.com unveils Matterport 3D Exterior	Industry deployment	CoStar/Apartments.com ships drone-captured Gaussian-splat building exteriors (PlayCanvas SuperSplat-style viewer) to millions of renters — a major non-tech public company standardizing on radiance fields, following Zillow.	美国 Apartments.com(CoStar)用无人机采集 + Gaussian Splatting 上线楼宇外景,面向数百万租房者,是科技圈外大型上市公司将辐射场标准化的标志(继 Zillow 之后)。
Domain deploys Gaussian-splat walkthroughs across 1700+ AU suburbs	Industry deployment	Australia's Domain (on CoStar's Matterport 3D Exteriors) rolls out 3DGS "dollhouse" walkthroughs across 1700+ suburbs with drone exteriors, virtual furniture removal, and measurements — evidence that 3DGS capture cost has dropped to mass-market scale.	澳洲 Domain(基于 CoStar Matterport 3D Exteriors)在 1700+ 城区铺开 3DGS 真实感"dollhouse"漫游(无人机外景、虚拟去家具、测距),表明 3DGS 采集成本已降至可规模化部署。

Datasets and Benchmarks

Resource	Type	Why It Is Useful	中文说明
Mip-NeRF 360 Dataset	Benchmark dataset	Common benchmark for unbounded 360-degree scenes.	常用 360 度无界场景基准数据。
Tanks and Temples	Benchmark dataset	Standard 3D reconstruction benchmark for intermediate and advanced scenes.	经典三维重建基准数据集。
KITTI-360	Urban driving dataset	Street-level urban data with camera, LiDAR, and annotations.	城市街景数据，包含相机、LiDAR 和标注。
Waymo Open Dataset	Autonomous driving dataset	Large-scale multimodal urban driving dataset.	大规模自动驾驶多模态城市数据。
OpenAerialMap	Aerial imagery	Public aerial imagery source for geospatial experiments.	可用于地理空间实验的公开航拍影像资源。
xBD	Disaster damage dataset	Building damage assessment dataset for disaster mapping research.	灾害建筑损伤评估数据集，适合和 3D 重建结果结合。

GeoAI and Disaster Mapping Ideas

English

• Post-disaster 3D inspection: reconstruct damaged neighborhoods from drone images and inspect structures in an interactive scene.
• Splat-to-map comparison: compare a Gaussian splat with orthomosaics, LiDAR point clouds, building footprints, and damage labels.
• Field evidence archive: store phone-based splats from field teams as spatially grounded qualitative evidence.
• Damage annotation support: use splat viewers to help annotators inspect facades, roofs, debris, and flood traces.
• Geospatial alignment: test how well splats can be georeferenced with GPS, visual positioning, ground control points, or LiDAR.
• Communication and outreach: publish lightweight 3D scenes for emergency management, public education, and research communication.

中文

• 灾后三维检查：使用无人机影像重建受灾社区，在交互式三维场景中检查建筑和道路状态。
• 三维结果与地图产品对比：将 Gaussian splat 与正射影像、LiDAR 点云、建筑轮廓和灾损标签进行比较。
• 现场证据归档：把现场团队用手机采集的 splat 作为带空间上下文的定性证据保存。
• 辅助灾损标注：使用 splat viewer 辅助标注人员检查立面、屋顶、碎片和洪水痕迹。
• 地理配准：测试 splat 与 GPS、VPS、地面控制点或 LiDAR 的对齐精度。
• 科学传播：发布轻量级三维场景，用于应急管理、公众教育和研究展示。

Suggested Experiments

Experiment	Input	Output	Evaluation
Phone scan pilot	Scaniverse or Polycam capture from a small outdoor site	Gaussian splat and mesh	Visual completeness, scale drift, upload and viewer quality
Drone image pilot	100 to 500 nadir and oblique drone images	COLMAP poses and 3DGS scene	Comparison with orthomosaic and sparse point cloud
Damage-scene pilot	Disaster imagery with building damage labels	3DGS scene plus label overlay	Whether 3D viewing helps identify damage evidence
Urban block pilot	Street-level video or image sequence	Large outdoor splat scene	Rendering quality, dynamic object artifacts, georeferencing error

Repository Roadmap

• Add installation notes for COLMAP, Nerfstudio, gsplat, and OpenSplat.
• Add a sample workflow for drone imagery: COLMAP to Nerfstudio Splatfacto to SuperSplat.
• Add a sample workflow for phone capture: Scaniverse or Polycam to web viewer.
• Add a disaster-mapping case template for documenting data source, event, location, reconstruction settings, and evaluation notes.
• Add a bibliography for 3DGS, NeRF, photogrammetry, visual localization, and GeoAI.

Citation

If this list helps your work, please cite the original tools and papers directly. This repository is a curated pointer list, not a replacement for the official documentation or publications.

如果这个资源库对你的工作有帮助，请优先引用原始工具、论文和官方文档。本仓库只是一个整理入口，不替代原始文献和项目说明。

License

This curated list is released under the Creative Commons Attribution 4.0 International License. Individual tools, datasets, and linked projects keep their own licenses.

本资源清单使用 Creative Commons Attribution 4.0 International License 发布。各工具、数据集和链接项目保留其各自的许可证。