What type of control data should be used to train YOLOv8

[limengqigithub]

Hello author, OmniParser is a good idea. I have a question to ask now. According to my understanding, YOLOv8 is trained using purely icon based control data, right? Control data with text did not participate in training?

[yadong-lu]

Hi, thanks for your interest. It is trained with interactive elements from the DOM tree using webpage data. So it can contain some of the text, e.g. hyperlinks etc