ncnn-swift

Examples on using ncnn with Swift.

ncnn is a really popular neural network inference framework optimzied for mobile platforms. Sadly, as for now there are few iOS examples and the existing ones (e.g. dangbo/ncnn-mobile, cmdbug/YOLOv5_NCNN) are mainly in Objective-C. In fact, Swift has gradually grown as the major prgramming language for iOS. As Apple releasing SwiftUI, a pure Swift UI framework, in WWDC2019, we could believe that most iOS developers will shift to Swift. Therefore, it is important for a mobile inference framework to support Swift.

And here we are. This project is include examples on how to connect ncnn framework with Swift.

SqueezenetSwift

This is a simple image classification with squeezenet from nihui/ncnn-android-squeezenet. I used bridge header to connect C++ and Swift. For an introduction of bridge header, please check this nice blog.

Yolov5Swift

This is a simple object detection example with yolov5 from nihui/ncnn-android-yolov5. It provides some ideas on how to register custom layer with Swift and how to use nihui/opencv-mobile together with ncnn.

Notice that the time shown includes cv::imread (around 40 ms) and the image is resized so that the longer edge is 640.

If you have any question on this project, please feel free to file an issue or PR.