MLPCondition

CLASS cleandiffuser.nn_diffusion.MLPCondition(in_dim: int, out_dim: int, hidden_dims: List[int], act: nn.Module = nn.LeakyReLU(), dropout: float = 0.25) [SOURCE]

A multi-layer perceptron (MLP) condition processor. The raw context tensor \(\bar{\bm c}\) is passed through a MLP to get the output tensor \(\bm c\).

Parameters:

• in_dim (int): The dimension of the input tensor \(\bar{\bm c}\).
• out_dim (int): The dimension of the output tensor \(\bm c\).
• hidden_dims (List[int]): The dimensions of the hidden layers of the MLP.
• act (nn.Module): The activation function of the hidden layers. Default is nn.LeakyReLU().
• dropout (float): The label dropout rate for the context tensor. Default is 0.25.

forward(condition: torch.Tensor, mask: Optional[torch.Tensor] = None) -> torch.Tensor

Parameters:

• condition (torch.Tensor): The context tensor in shape (b, ..., in_dim).
• mask (Optional[torch.Tensor]): The mask tensor. Default is None. None means no mask.

Returns:

• torch.Tensor: The output tensor in shape (b, ..., out_dim). Each element in the batch has a probability of dropout to be zeros.