Issue with diffusion objective: Code predicts noise while Table 5 suggests predicting $x_0$

[Iuek]

The paper states that the model predicts the true value ($x_0$), but the provided source code actually predicts the noise ($\epsilon$). Is there an issue with this baseline implementation?