All notable changes to this project are documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Removed option to create top level of every octave from input image PR
- Removed option to compute very narrow Gaussian filters called opencv PR
- Removed unreachable code for alternative downscaling between octaves PR
- Support for device selection and multiple GPUs PR
- Support for CUDA versions through 12.1 PR
- GitHub Actions for CI PR
- Missing thrust include PR
- Required thrust include for
s_filtergrid.cuPR - Added vlFeat's descriptor extraction method as an option PR
- Added CI for windows on Github Actions and refactoring PR
- CMake: CUDA as first-order language, different CC selection PR
- CMake: SM 86 only for CUDA >= 11.1 PR
- CMake: Suppress deprecated CUDA SM warnings PR
- Replace exit() calls with thrown exceptions PR
- Errors with large images PR
- CMake: Explicit CC list for Jetson and Tegra platforms PR
- Remove broken and unused code path from L2 normalization PR
- Do not use NVTX PR
- Documentation: Biblio build PR
- Fixed include for CUDA 13 PR
- Update with vcpkg package PR
- Support for CUDA 11 PR
- Improved checks for CUDA textures PR
- Documentation: ReadTheDocs integration PR
- CMake: Support for SM86 compute capability PR
- Race conditions identified by racecheck PR
- CMake: Windows build fixes PR
- Memory management: pair
mallocwithfree, notdeletePR - CUDA: Pass correct shared memory size to orientation kernel PR
- Testing paths fixes PR
- Remove stale undefined function PR
- Remove boost dependency from core popsift library PR
- Bugfix: Support for images with different resolutions PR
- CMake: Auto-build export symbols for shared libs on Windows PR
- Bugfix: freeing page-aligned memory on Win32 PR
- Paper published @MMSys18 (https://dl.acm.org/doi/10.1145/3204949.3208136)
- Docker support PR
- Appveyor CI windows PR
- Support for Cuda 9 PR
- Thrust with Cuda 7 PR