Skip to content

Add Padding as an Argument to AutoEncoder #7045

Closed

Description

@surajpaib

Is your feature request related to a problem? Please describe.
In the current implementation of the AutoEncoder class in the monai.networks.nets module, the padding parameter is not customizable. This limits the model's flexibility, as different padding strategies might suit different data-sizes.

Describe the solution you'd like
I would like to propose adding a padding argument to the AutoEncoder class constructor. This would allow users to specify the padding strategy when instantiating the model.

Describe alternatives you've considered
An alternative solution could be to create a subclass of AutoEncoder (like CustomAE) whenever a different padding strategy is needed. See below,

class CustomAE(AutoEncoder):
    def __init__(self, padding, **kwargs):
        self.padding = padding
        super().__init__(**kwargs)
    
    def _get_encode_layer(self, in_channels: int, out_channels: int, strides: int, is_last: bool) -> nn.Module:
        # ...
        mod = Convolution(
            spatial_dims=self.dimensions,
            in_channels=in_channels,
            out_channels=out_channels,
            strides=strides,
            kernel_size=self.kernel_size,
            padding=self.padding,
            # ...
        )
        return mod

Activity

  1. added a commit that references this issue on Sep 30, 2023
    2b39067
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions