Releases: zlogic/cybervision
Releases · zlogic/cybervision
Rust version 0.14.0
- Significantly improved performance of high performance GPUs (M1, Geforce RTX)
- Added a low-power GPU mode to prevent device lost errors in integrated Intel GPUs
- Fixed fundamental matric in perspective projection - resulting in a significant improvement in quality
- For OBJ and PLY files, it's possible to assign point colors from the source image
- For OBJ files, it's now possible to use texture from the source image
Rust version 0.13.2
Improved GPU performance.
Rust version 0.13.1
Fixed a few performance issues:
- Unnecessary calls to GPU
- Writing *.ply and *.obj files is significantly faster
Rust version 0.13.0
Rewrote code from C to Rust. This rewrite helped discovering and fixing several bugs, and also improved maintainability:
- Replaced CLAPACK/OpenBLAS (Windows and Linux) and Accelerate (macOS) with nalgebra.
- This removes tricky C/ASM/Fortran code that's impossible to troubleshoot.
- Most loops are now unrolled, and there are almost no conversions when multiplying matrices.
- Inlining functions also improves performance.
- Rewrote FAST into Rust, in a parallelized way (slower than the C version, but runs faster on multi-core CPUs).
- Replaced Vulkan/Metal with wgpu; the same codebase can target Vulkan, Metal or DirectX 12 - whatever is available.
- Almost all Rust libraries are provided in source form, so the compiler/linker can exclude unused code.
C version 0.12.0
Fixed a few bugs that were discovered during the rewrite into Rust.
C version 0.11.0
- Significantly improved accuracy of perspective reconstruction by using the "Gold standard" algorithm.
- Remove noisy data from images with perspective projection.
C version 0.10.0
A few tweaks and precision improvements:
- Histogram filter to remove peaks - greatly improves results for SEM images
- A few (failed) attempts to reconstruct photos with perspective projection
- Perspective projection is a lot more sensitive to noise and requires additional steps to recover fundamental matrix
C version 0.9.0
Changed the way correlation works - now using a fundamental matrix, which significantly improves accuracy and also allows reconstruction of normal photos (using perspective projection instead of parallel projection).
- Fixed a bug where number of point matches is incorrectly counted
- Fixed colormap rendering in PNG images
- Changed colormap to Viridis
With this release, images 2 and 3 fail to be reconstructed - because the point matching and accuracy standards have been raised.
However now most photos produce more realistic results, and image 05 can now be reconstructed.
C version 0.8.1
- Removed IDW interpolation (slow, unsatisfactory results)
- Switched to PThreads4W in Windows (it's used by other libraries and is already included anyway)
C version 0.8.0
- Significantly improved CPU-only performance
- Added option to use Inverse Distance Weighting interpolation
- Allow to output results without interpolation