Implemented in pure pascal LightNet is an artificial intelligence neural network library Inspired by Darknet and YOLOv7 github C library, this implementation can run most of the darknet including YOLO models nativly and self dependently no external libraries required.
- Currently training & inference runs on CPU , however i have notice that inference runs faster than the Optimized C CPU implementation
- Optionally, using
-dOPENBLASor-dMKL( Custom options in Lazarus-Ide for example ) will enable "libopenblas" or "Intel's Math Kernel Library" which is further optimized for CPU, however, on lazarus no significant improvement was noticed. - Before running the test make sure that yolov7.weights file is placed next to the test project executable.
- Other .weights models should work (check "cfg" folder)
- More Examples coming soon
- OpenCL and CUDA implementation
- Test on other platforms (MacOS, Linux)
- Test on Single Board Computers (NVIDIA Jetson, Raspberry PI, etc..)
tests provided for both Lazarus and Delphi
this project is still at the early stage, if you wish to see more models running in pure pascal, please consider donating here
