A problem in labeling image data

[ghost]

Hi, experiencor, i'm using yolov3 model to detect people who is using mobilephone in a working scene. There are people working in front of the computer and people walking around in that scene. The action of using phone is usually a small target and sometimes covered partly by some other things. The approach what i take to label the image data seems not work well . So can you give some advice on how to label the image data to achive a good detection result.

[darjoo]

If you could show some examples of the scene and how you have been labeling, we would be able to help you much better then. :)