Skip to content

Hard-negative mining using __getitem__ directive in Dataset class #2407

@yannadani

Description

@yannadani

If I wish to do online hard negative mining (for triplet based algorithms), then I wish to send the recent state dict of my network to the data loader to evaluate the criterion and select the negative. Currently, you need to have a lot of workaround for this to get it working. I think the major limitation is the usage of the getitem directive. Can we have a more flexible approach which allows to send arbitrary number of parameters? Or is there a solution in the current situation which I am not aware of?

Thanks.

cc @ssnl

Metadata

Metadata

Assignees

No one assigned

    Labels

    module: dataloaderRelated to torch.utils.data.DataLoader and SamplertriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions