Our take
Inactiveno commits in over a year- Activity
- 19
- Community
- 4
- Issues
- 21
- Pull requests
- 20
+2 stars in the last 30 days
- Fewer contributors than 91% of the projects we track
GigaSLAM presents a research-focused approach to large-scale monocular outdoor SLAM, but its inactive, single-contributor state makes adoption dependent on a team’s ability to maintain and troubleshoot the stack.
Good fit if
- You are researching monocular SLAM or Gaussian-splat mapping on the listed outdoor benchmarks.
- You can build and support a Linux, CUDA, and C++ environment, including the project’s rendering and loop-closure modules.
- You are comfortable working from an MIT-licensed research repository with no recent maintenance activity.
Look elsewhere if
- You need a currently maintained project: there have been no commits in over a year, no releases in the past year, and most of the code comes from one person.
- Your deployment needs CI, automated tests, or evidence of recent issue and pull-request handling, which the repository does not report.
- You need a lightweight setup or cannot provide a CUDA-capable NVIDIA GPU and substantial CPU memory for long sequences.
All health signals
| Last commit | 2025-06-12 (1.3 years ago) |
|---|---|
| Commits, last 90 days | 0 |
| Releases, last 12 months | 0 |
| Contributors | 1 |
| Issues closed, last 90 days | 0 |
| Pull requests merged, last 90 days | 0 |
| Project age | 19 months |
Checked on 2026-10-15 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
151Forks
13Change since Nov 17, 2026: 0 stars·+1 forks
Overview
GigaSLAM is a monocular RGB SLAM system for kilometer-scale outdoor environments, built on Scaffold-GS and the MonoGS framework. It addresses the difficulty of tracking and reconstructing expansive, unbounded scenes, where many NeRF and 3D Gaussian Splatting SLAM systems are geared toward smaller indoor settings. The authors demonstrate it on KITTI, KITTI 360, 4 Seasons, and A2D2.
Its mapping uses a hierarchical sparse voxel representation with Gaussians decoded at multiple levels of detail. For tracking, it combines metric depth from UniDepth with epipolar geometry and PnP, and uses Bag-of-Words loop closure to help maintain alignment over long trajectories. The documented setup targets Linux with CUDA and a CUDA-capable NVIDIA GPU.
Key Features
- Builds a sparse, hierarchical voxel map for large outdoor scenes.
- Decodes Gaussian splats at multiple levels of detail for scene rendering.
- Uses UniDepth, epipolar geometry, and PnP for monocular pose estimation.
- Adds Bag-of-Words loop closure to correct alignment over long runs.
- Includes configurations and demonstrations for KITTI, KITTI 360, 4 Seasons, and A2D2.
- Provides optional visualization of SLAM results during execution.
Use Cases
- Researchers comparing monocular SLAM approaches on urban outdoor datasets.
- Robotics and autonomous-driving teams exploring large-scene mapping from RGB camera sequences.
- 3D vision developers studying Gaussian-splat rendering within a SLAM pipeline.
- Practitioners who need to inspect reconstructed scenes and tracking results across long trajectories.
What you need
Detected in the repository
- Python (from setup.py)
License in plain words
MITpermissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the copyright and license notice
- Share your changes: no
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
Follow the README for environment setup, CUDA/C++ module compilation, model weights, and dataset configuration. After configuring a dataset and camera intrinsics, run:
python slam.py --config ./configs/kitti_06.yaml
Considerations
The documented environment uses Ubuntu 22.04, CUDA 11.8, PyTorch 2.2, and an NVIDIA GPU. Setup also involves compiling CUDA/C++ components, installing OpenCV and DBoW2, and downloading pretrained model weights. The README recommends a server with at least 32 GiB of CPU RAM; it reports that longer sequences may require more memory. Dependency versions, especially the coupling between xFormers, PyTorch, and CUDA, require care. The repository has no reported CI or tests, and its last commit was over a year ago, so compatibility with newer dependencies and ongoing support are uncertain.
Found this useful?
Share it with someone who would like GigaSLAM.