pySLAM v.2.10.0 – A Hybrid Python/C++ Framework for Visual SLAM and 3D Perception
I’m happy to share pySLAM v2.10.0 🎉
🔗 https://github.com/luigifreda/pyslam
This is a major update that moves the project forward on multiple fronts:
• Runtime performance
• Semantic segmentation tightly integrated with volumetric mapping
• 3D scene inference
pySLAM is now a hybrid Python/C++ Visual SLAM framework supporting monocular, stereo, and RGB-D pipelines, bringing together local features, loop closing, dense mapping, depth prediction, and semantic understanding in a single environment.
📌 New modular C++ SLAM core
This release introduces a modular sparse SLAM core implemented in both Python and C++, exposed via pybind11. You’re free to use either implementation interchangeably: maps saved by one can be loaded and continued by the other.
The key outcome is simple: run the full pipeline at the native tempo of standard datasets with the C++ core, while retaining the original Python core where it shines – configuration, integration, and rapid iteration.
📌 Expanded semantic segmentation and semantic volumetric mapping
v2.10.0 significantly extends semantic segmentation support, integrating modern models such as YOLO, RFDETR, DETIC, ODISE, and EOV-SEG directly into the SLAM pipeline.
Semantic predictions are incrementally fused into volumetric map representations, enabling:
• Semantic voxel grids with class-aware color maps
• 3D object-level segmentation inside volumetric reconstructions
• Joint reasoning over geometry, appearance, and semantics in 3D
Semantic information is now a first-class component of both sparse and volumetric maps.
📌 New end-to-end multi-view 3D scene inference pipeline
v2.10.0 adds a unified and modular pipeline for end-to-end inference of 3D scenes from multiple images, with support for: DUSt3R / Mast3r / MV-DUSt3R, VGGT (and Robust VGGT), DepthFromAnythingV3, and Fast3R.
📌 Plus: datasets, depth models, and visualization
The release also includes upgrades across volumetric reconstruction, depth prediction models, dataset support, ROS1 / ROS2 handling (including a new MCAP player), and visualization tools.
