Skip to content

Proposal of modification of ComposeMany for pre-split tensors #18

Description

@Pliploop

Hello,

I'm using your library for a project and have come across a use-case which is not covered, I'd like to implement it and merge it if that's alright.

I'm currently pre-splitting an audio tensor into a shape [N_augmentations, T], where T is the number of samples. This pre-split is important as it is the consequence of a sampling strategy. Currently this library only supports augmentation into multiple views of one sample sample or the same augmentations over a batch.

I'd like to propose a modification of the ComposeMany class to support pre-split tensors. i.e shape [N,T] going in and [N,T] coming out, but with different augmentations for each n in range(N)

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