Welcome to the InternLandMark homepage. InternLandMark is mainly developed by Shanghai AI Laboratory. We welcome contributions to our project in different forms.

We provide 🔥CityEyes🔥, which is a public experiential project where everyone can freely travel and edit the city scenes provided within it.

If you want to learn more about us, please click here.

🔥 LandMark 3.0 [Doc]

• [08/2024] We public the LandMarkSystem, which is the first open-source system for large-scale scene reconstruction training and real-time rendering.

• [08/2024] We develop the FlashGS to enable real-time 3D Gaussian Splatting (3DGS) based rendering especially for large-scale and high-resolution scenes.

⭐ LandMark 2.0 [Doc]

• [06/2024] We expand the model's editable capabilities and have launched a public experiential project, called CityEyes, where everyone can freely travel and edit the city scenes.

• [03/2024] Inspired by the Level-of-Detail (LOD) techniques, we introduce Octree-GS, featuring an LOD-structured 3D Gaussian approach supporting level-of-detail decomposition for scene representation that contributes to the large scenes rendering.

• [11/2023] We propose the Scaffold-GS (CVPR24, Highlight), which combines the high-performance rendering efficiency of 3D Gaussian Splatting with the flexibility and high quality of various classic NeRF representations.

🎉 LandMark 1.0 [Doc]

• [01/2024] We release code about kernel optimization of dynamic fetching rendering in LandMark.

• [11/2023] We release code about dynamic fetching rendering in LandMark.

• [09/2023] We release code about hybrid parallel training with model parallel and DDP training in LandMark.

• [09/2023] We build a large-scale, comprehensive, and high-quality synthetic dataset MatrixCity (ICCV 2023) for city-scale neural rendering researches.

• [07/2023] We propose the project LandMark, the groundbreaking large-scale 3D real-world city scene modeling and rendering system. The project is built upon GridNeRF (CVPR23).

• GridNerf: Grid-guided Neural Radiance Fields for Large Urban Scenes.
• Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering.
• Octree-GS: Towards Consistent Real-time Rendering with LOD-Structured 3D Gaussians.

• MatrixCity: A Large-scale City Dataset for City-scale Neural Rendering and Beyond.

• LandMarkSystem: The first open-source system for large-scale scene reconstruction training and real-time rendering.
• FlashGS: An efficient CUDA Python library, enabling real-time 3D Gaussian Splatting (3DGS) based rendering especially for large-scale and high-resolution scenes.