This is a C++ inference implementation of Yolov5 with EdgeTPU. Runs at 12 FPS on desktop with Coral EdgeTPU enabled.
First you will need absl library to be installed, refer to quickstart to install absl library
From the project directory, run the following command:
create build directory to keep source directory clean: mkdir build && cd build
generate makefile: cmake ..
compile: make -j4
The executable file will be located at out
folder in project folder. Run the demo with command out/k8/Yolo_EdgeTPU --model_path=MODEL_PATH --label_path=LABEL_PATH
from project folder where model_path and label_path should point to the files in model
directory.
Thanks to excellent yolov5 work from Ultralytics and model conversion code from zldrobit