Questions about details of t2m evaluation and release of baseline script

[PerfectBlueFeynman]

Dear authors,

Thank you for the amazing work on the new, rich motion-text datasets.

I'm trying to reproduce the evaluation results using Motion-X on several models you listed in the paper.

May I ask how you handle frame-level pose text with word and pos tagging in t2m text encoder?
Do you use posetext or semantic label for evaluation?
For cross-dataset comparison with HumanML3D, do you leave HumanML3D to 20 fps or matching the fps of Motion-x and humanml3d (30 vs 20 fps)?
How do you setup the parameters for t2m matching model training on Motion-X?
Why is the testing results on Humanml3d with MLD trained on Humanml3d is different than their original paper?

And you mentioned in previous issues (#33 (comment)) that you're about to release the baseline script, may I ask the estimated date for this?

Thank you!