Inquery about table 4 in paper and preprocessng to one annotation format method

Hi, Thank you for sharing this amazing work and for the detailed paper. I have a few questions and would really appreciate your insights: 1. Regarding the preprocessing of different annotations into a unified format, could you provide more details? Specifically, are you aware of any existing solutions for converting different formats to ARKit? (In the paper, the conversion is to FLAME.) Additionally, would you be willing to share the code used for this preprocessing? 2. In Table 4, you compare preprocessing (one decoder) and UniTalker in terms of LVE, which does not account for full facial movement. Have you also evaluated the results on other metrics such as MVE, UFVE, and FDD? If so, I would love to hear any insights you can share. Looking forward to your response. Thanks in advance!