Questions about retraining MoST using 100STYLE dataset

[Dandrof]

Hello, I have retargetted the 100STYLE dataset to the skeleton of the xia dataset, and then I selected 8 styles from it and retrained the model following the process outlined in the README. However, the results are not ideal; for example, it fails to capture the Aeroplane style. I modified the contents of xia_dataset.yml. Could there be any code changes that I missed?

[Boeun-Kim]

Thank you for your interest in our research. Assuming that you have retargeted the 100Style dataset to align all joint positions with the XIA dataset, it seems that the code related to body structure does not need to be modified (e.g., model/networks.py). However, as dataset characteristics can differ, hyperparameters should be adjusted empirically (arguments.py). Additionally, it may be helpful to review the data preprocessing part (data/xia_preprocess.py).