Method forward

skills/paper2code/worked/ddpm/src/loss.py:43–60

§3.4, Eq. 14 — L_simple = E[||ε − ε_θ(x_t, t)||²] Args: noise_pred: (batch, C, H, W) — predicted noise ε_θ(x_t, t) noise_true: (batch, C, H, W) — true noise ε ~ N(0, I) Returns: scalar — mean squared error loss

(
        self,
        noise_pred: torch.Tensor,
        noise_true: torch.Tensor,
    )

        super().__init__()

    def forward(
        self,
        noise_pred: torch.Tensor,
        noise_true: torch.Tensor,
    ) -> torch.Tensor:
        """
        §3.4, Eq. 14 — L_simple = E[||ε − ε_θ(x_t, t)||²]

        Args:
            noise_pred: (batch, C, H, W) — predicted noise ε_θ(x_t, t)
            noise_true: (batch, C, H, W) — true noise ε ~ N(0, I)

        Returns:
            scalar — mean squared error loss
        """
        # §3.4 — Simple MSE between predicted and true noise
        # "equivalent to (a re-weighted variant of) the ELBO"
        return nn.functional.mse_loss(noise_pred, noise_true)