Hello,I have been studying the Stochastic Centering mechanism described in the RFdiffusion2 supplementary.The text mentions that the global translation $\epsilon$ is sampled from $\mathcal{N}(0, I_3 \cdot \sigma_{perturb})$. However, I could not find the exact numerical value (or range) used for this hyperparameter in the paper or the provided snippets. Could you please clarify the specific value of $\sigma_{perturb}$ (in Angstroms) used during your training experiments?
Thank you for your amazing work!