Skip to content

How to evaluate naive UI Agents which predict box/point directly? #12

Description

@zhangxgu

Hi,
it's a nice work.

However, we find that the official code of WebVoyager is to use som image and list the elements in order as input, then requiring the model to output the index. However, many naive UI Agents are trained to output the box/point directly. Thus, we wonder to know how to map the outputs including actions and points to the WebElements?
Thanks!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions