BarlowTwins

BarlowTwins#

class stable_pretraining.methods.BarlowTwins(encoder_name: str | Module = 'vit_small_patch16_224', projector_dims: Sequence[int] = (8192, 8192, 8192), lambd: float = 0.0051, low_resolution: bool = False, pretrained: bool = False)[source]#

Bases: Module

Barlow Twins self-supervised learning.

Parameters:

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

  • projector_dims – Hidden + output dims (default (8192, 8192, 8192) matches the ResNet50 recipe).

  • lambd – Off-diagonal weight in the cross-correlation loss (default 5.1e-3 from the paper).

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

  • pretrained – Load pretrained timm weights.

forward(view1: Tensor, view2: Tensor | None = None) → BarlowTwinsOutput[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