where I can find corresponding code about Motion data augmentation?

[miaoYuanyuan]

Based on my own understand, there are 3 parts process about traing.

1. Randomized Shuffled Anchor Pose: corresponding to the random mask_start_frame.
2. Semantic Embedding: in the network Sturcture, cond_embedding
3. motion data augmentation? I can't find the corresponding code?

[jihoonerd]

@miaoYuanyuan Our motion data augmentation is based on the paper and this is used to generate more diverse root trajectory from existing motion. But the algorithm is private and implemented in MATALB, so we haven't integrated it into this codebase. But I believe current published version will be able to perform if given pose is generalizable from the training dataset.