You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for your sharing.
My request is like:
They tried a lost of backbones in original Diffusion Policy, which includes the transformer. Even in their paper, they claimed that the transformer seems to be not as good and stable as Unet backbone, but from my previous reproduce and others work, it is not quite true to claim transformer is worse (no matter the number of params).
Hence, I suggest to add the transformer support in lerobot project. As an advice: You can add a Class named DiffusionTransformer in modeling_diffusion.py and let the user to select the backbone in configuration file.
The text was updated successfully, but these errors were encountered:
Thanks for your sharing.
My request is like:
They tried a lost of backbones in original Diffusion Policy, which includes the transformer. Even in their paper, they claimed that the transformer seems to be not as good and stable as Unet backbone, but from my previous reproduce and others work, it is not quite true to claim transformer is worse (no matter the number of params).
Hence, I suggest to add the transformer support in lerobot project. As an advice: You can add a Class named DiffusionTransformer in modeling_diffusion.py and let the user to select the backbone in configuration file.
The text was updated successfully, but these errors were encountered: