How smartroboticslab/okvis2 works
Project type and purpose
• This is OKVIS2: an open-source, modular C++ library and set of applications for real-time keyframe-based visual-inertial SLAM (Simultaneous Localization and Mapping).
• It comprises low-level utilities, sensor-handling, frontend feature pipelines, backend optimization, and ROS2 integration, all built via CMake.
File structure
• Top-level folders:
• cmake / CMakeLists.txt → build configuration
• external → third-party submodules (DBoW2, opengv, Ceres, etc.)
• config / launch → YAML configs and ROS2 launch files
• okvis_apps → demo & recorder executables
• okvis_common, okvis_cv, okvis_frontend, okvis_ceres, okvis_kinematics, okvis_multisensor_processing, okvis_ros2, okvis_time, okvis_timing, okvis_util → core library modules
• resources / cnn → segmentation model artifacts and helper scripts
• Pattern: layered, modular libraries with clear dependency flows (common → cv → frontend → ceres backend; multisensor processing → frontend/backend; ros2 wrappers on top).
README for additional insights
• Confirms pure CMake, key dependencies (Eigen, Ceres, Glog, OpenCV, LibTorch for optional CNN), ROS2 build option.
• Describes pipeline callbacks (optimised graph & IMU callbacks) and trajectory management.
• Describes demo apps and ROS2 nodes for live processing or bag playback.