- [2024/07]: We release LandMarkSystem, which is the first open-source system for large-scale scene reconstruction training and rendering.
LandMarkSystem provides a multi-algorithms system for 3D reconstruction tasks in any scale, offering a high degree of flexibility and efficiency through its modular design and advanced parallelization strategies. This system has the potential to significantly advance the state of the art in large-scale scene reconstruction by enabling the deployment of NeRF and 3DGS algorithms in a manner that is both scalable and resource-efficient.
| Model | Training | Rendering | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Single GPU | Model Parallel | Data Parallel | Dynamic Loading | Single GPU | Model Parallel | Data Parallel | Dynamic Loading | Kernel Optimization | |
| GridNeRF | ✅ | ✅ | ✅ | - | ✅ | ✅ | ✅ | ✅ | - |
| InstantNGP | ✅ | - | ✅ | - | ✅ | - | ✅ | ✅ | - |
| Nerfacto | ✅ | - | ✅ | - | ✅ | - | ✅ | ✅ | - |
| Vanilla GS | ✅ | - | ⌛ | ⌛ | ✅ | - | ⌛ | ⌛ | ✅ |
| Scaffold GS | ✅ | - | ⌛ | ⌛ | ✅ | - | ⌛ | ⌛ | ✅ |
| Octree GS | ✅ | - | ⌛ | ⌛ | ✅ | - | ⌛ | ⌛ | ✅ |
In the table above, ✅ indicates that the feature is supported in the current version, and ⌛ indicates that the feature will be supported in the next open-source release.
Please refer to Usage to start installation, training and rendering.
For more details, please check Documentation.
For more information about the system architecture and training/rendering performance, please refer to our technical report (TBD).
The main work comes from the LandMark Team, Shanghai AI Laboratory.
