Draw object detection box at high speed [10 times faster version]

Draws object detection results at high speed

Drawing becomes a bottleneck rather than detection

The object detection model for mobile such as Yolo can be executed on iPhone 11 in the order of 0.02, but if you draw the result on UIImage by the method like my article below, the drawing process will take about 0.5 seconds.

Visualize the object detection box Draw a bounding box on an image on iOS

In other words, the
drawing process took 25 times longer than the detection itself.
In particular, drawing label text consumes 80% of the processing.

This is fatal for object detection processing of a large number of frames such as moving images .

10 times faster if processed with CGImage

If you draw on CGImage using CGContext, you can draw in 0.04 seconds in the same environment, which
is more than 10 times faster than drawing on UIImage.

Yolov5 detection and drawing sample code

GitHub - john-rocky/CoreML-YOLOv5: A sample project how to use YOLOv5 in iOS

🐣

I’m a freelance engineer.
Work consultation
Please feel free to contact us with a brief development description.
rockyshikoku@gmail.com

I am making an app that uses Core ML and ARKit.
We send machine learning / AR related information.

GitHub

Twitter
Medium