Skip to content

question for your designed model details #8

@huinee56

Description

@huinee56

In your paper, section 3.4 "Recursive Joint Cross-Modal Attention", you mentioned" the use of a fully connected layer in equation (1). And, in equations (2\3\4) and (5\6\7) , the learnable weight matrix appears to be inconsistent with the specific implementation details that I found in the DCNLayer class code you provided. Additionally, in your paper, you describe that "In order to obtain more refined feature representations,
the attended features of each modality are again fed as input to the respective joint cross-modal attention module," referring to equation (11), but I cannot find a corresponding implementation for this. Could you kindly clarify this issue? If possible, could you provide a more detailed related code?

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