WMSE

Contents

WMSE#

class stable_pretraining.methods.WMSE(encoder_name: str | Module = 'vit_small_patch16_224', projector_dims: Sequence[int] = (1024, 64), eps: float = 0.001, low_resolution: bool = False, pretrained: bool = False)[source]#

Bases: Module

W-MSE: whitening + MSE between paired views.

Parameters:

  • encoder_name – timm model name or pre-built nn.Module.

  • projector_dims – (hidden, output) for the projector (default (1024, 64); a small whitening dim helps stability).

  • eps – Cholesky regularisation (default 1e-3).

  • low_resolution – Adapt first conv for low-res input.

  • pretrained – Load pretrained timm weights.

forward(view1: Tensor, view2: Tensor | None = None) → WMSEOutput[source]#

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

Returns:

Your model’s output

Contents