Releases: uos/rmagine
Releases · uos/rmagine
v2.2.8
This is a new version of rmagine, which migrated some Mesh ICP computations from RMCL. Several tests have been added, I also took the opportunity to clean up the interface for single hypothesis correction aka pose tracking. All other external software should still work, so I just increased the patch version.
New
- CrossStatistics math type, which contains two means a covariance and the number of measurements. Also the reduction function from the MICP-L paper is integrated
- PointCloudView. Contains views to existing memory buffer. We use it for cleaner, and more readable interfaces, while not losing any runtime.
- Functions for registration. Correspondence search, Umeyama-based optimization
Improvements
- integrated support for multi-block computation in the non-batch version of "sum"
Tests
- statistics for CPU (core)
- statistics for GPU (cuda)
- Embree correction: finding RCC using embree, optimizing using CPU-Umeyama
- OptiX correction: finding RCC using optix, optimizing using GPU-Umeyama
v2.2.7
v2.2.6
v2.2.6
- SVD implementation based on "Numerical Recipes" that works out of the box on CPU as well as in CUDA kernels. First test results show speed up and accuracy over Eigen/cusolver. Need to conduct further experiments. Perhaps directly with RMCL.
- added convenience functions to sensor models and embree simulators.
- Added rmagine-ouster component
- Dependency: jsoncpp
- Loading Ouster Meta files as rmagine sensor model
Additional things
- removed obsolote functions.
- made types polyscope compatible
v2.2.5
v2.2.4
v2.2.3
v2.2.2
v2.2.1
v2.2.0
Rework of build and install system.
Major changes:
- Components for each part of the library: rmagine::core, rmagine::embree, rmagine::cuda, rmagine::optix
- Debian package generation with CPack
- Removed Embree 3 support
Note: Attached binaries are compiled with Ubuntu 20.04, Embree 4.2.0, Cuda 11.4, Nvidia-driver 470.199.02, OptiX 7.3.0.
v2.1.1
Patch:
Preparations to fully implement new feature: Integrate CUDA unified memory.
- Implemented respective Memory object UNIFIED_CUDA including alloc, free, copy, stream copy functions
- OptiX simulators can now additionally simulate to unified memory. Mixing memory is possible as well.
TODOs missing to next minor version change: Add UNIFIED_CUDA into Embree Simulators without breaking compatibility for CPU only devices