Skip to content

[E2E evaluation] ObjectDetection and TextToBbox appear ambiguous in the released tool descriptions #5

Description

@StarryProgrammer

In my full end-to-end evaluation, the reference trajectories contain
ObjectDetection in 21 samples. However, the model selected ObjectDetection in
only 6 samples, 5 of which were executed successfully, and none overlapped with
those reference samples. In these cases, the model generally selected
TextToBbox instead.

Would it be possible to make the functional descriptions of ObjectDetection and
TextToBbox more explicit, so that the model can distinguish between them more
reliably? I wonder whether such a clarification could improve the evaluation
results.

Thank you very much for your work!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions