Skip to content

SMP train conditional prior #116

@guptadivyansh

Description

@guptadivyansh

As shown in the SMP paper, it would be super useful to train the diffusion model with a conditioning variable that can guide the diffusion model in downstream RL tasks.

I see that the TinyMDM model already supports CondDiT as an arch, but train_tinymdm.py does not route any conditioning class labels to it at the moment, partly because theMotionPriorData class does not load/store it.

Am I missing to see this somewhere? Otherwise, I can try to add a small PR to add this.

@Murrol

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions