FastYolo 11.1.1

Fast .NET Yolo library, requires Cuda 11.1+ to be installed to work, it will use the GPU for processing (which is many times faster than using the CPU version disabled in this release). Since v11.1 this library requires Cuda 11.1 and a GPU with shader model 7.5 (on Windows, e.g. RTX 20x0+) and 5.3 (Jetson Nano)+7.2 (Jetson Xavier) for half-float optimizations with even faster tensor core computations (which will not work on 1660 GPUs, which have no tensor cores, those will switch back to cuda cores, sm 7.5 as well), older GPUs with sm 5.2+ (9x0 series, 10x0 series and jetson nano) are not supported anymore, use FastYolo v1.0.4 for that. More details about Cuda Cores and Compute Capability: https://www.geeks3d.com/20100606/gpu-computing-nvidia-cuda-compute-capability-comparative-table/

Includes the .dll files for Yolo Darknet Wrapper, Real-Time Object Detection (AlexeyAB/darknet with additions and fixes here: https://github.com/maalik0786/darknet), including cudnn64_8.dll, opencv_world440.dll and pthreadVC2.dll as needed by the yolo_cpp_dll implementation. Tested on Windows x64, Linux AMD64 and ARM64 (Jetson/Raspi4). On Linux OpenCV and Cuda+CudNN must be installed already, libdarknet and libpthread are provided for both AMD64 and ARM64 inside the nuget package.

Since v11.1 the default resolution is now 640x480, which works a bit better for most pictures and camera data. v1.0.4 adds resize detection and will output a warning in the console and allow to check via the CheckIfImageWasResized method if any resize happened. Resizing should not be done dynamically in the darknet library, extra memory would be required and it is obviously much slower. Since 1.0.3 it also supports OpenCV for more advanced features like tracking, shape detection, etc.