Skip to content

Make dictionary transforms support different random parameters across keys #7561

Description

@surajpaib

Is your feature request related to a problem? Please describe.
For current dictionary transforms, the random parameters are shared across keys.

E.g: see

# all the keys share the same random shift params

self.randomize(first_item.peek_pending_shape() if isinstance(first_item, MetaTensor) else first_item.shape[1:])

Describe the solution you'd like
An option to have separate random parameters would be useful for non input-target dictionary structure. Eg: for dicts that contain views in self-supervised learning. An additional key to the random dict transforms would be ideal

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions