rknn-cpp-Multithreading
This open-source project builds upon the rknpu2, using a thread pool to enable asynchronous RKNN model operations on RK3588/RK3588s, thus improving NPU utilization and inference speeds. The project incorporates relu activation function optimization for yolov5s models. Recent changes have resolved cmake pthread issues, added a nosigmoid branch for better performance with rknn_model_zoo models, and updated the NPU SDK to version 1.5.0. The project supports multi-threaded performance testing with OpenCV, and provides guidance for application deployment using the rkYolov5s.hpp.
