Hi,
First of all, thank you so much for your excellent open-source work! It is very inspiring and helpful for my current research in motion generation.
I am currently trying to reproduce your results but have run into a few issues. I used the command provided in the repository on my single-GPU setup:
accelerate launch --config_file 1gpu.yaml --gpu_ids 0 -m scripts.train --name t2m_condunet1d --model-ema --dataset_name t2m
However, I haven't been able to fully reproduce the expected performance with this. Considering you might have performed hyperparameter tuning, could you please share your tuning methods or specific hyperparameter configurations? Furthermore, it would be greatly appreciated if you could provide the complete and exact training and evaluation commands you used for both the HumanML3D and KIT-ML datasets.
Additionally, to ensure that my data processing perfectly aligns with yours and to rule out any data-related issues, would it be possible for you to share your processed datasets directly (e.g., via a Google Drive link)?
Thank you again for your time and your valuable contribution to the community!
Best regards
Hi,
First of all, thank you so much for your excellent open-source work! It is very inspiring and helpful for my current research in motion generation.
I am currently trying to reproduce your results but have run into a few issues. I used the command provided in the repository on my single-GPU setup:
accelerate launch --config_file 1gpu.yaml --gpu_ids 0 -m scripts.train --name t2m_condunet1d --model-ema --dataset_name t2m
However, I haven't been able to fully reproduce the expected performance with this. Considering you might have performed hyperparameter tuning, could you please share your tuning methods or specific hyperparameter configurations? Furthermore, it would be greatly appreciated if you could provide the complete and exact training and evaluation commands you used for both the HumanML3D and KIT-ML datasets.
Additionally, to ensure that my data processing perfectly aligns with yours and to rule out any data-related issues, would it be possible for you to share your processed datasets directly (e.g., via a Google Drive link)?
Thank you again for your time and your valuable contribution to the community!
Best regards