Generate large datasets for model training

[wenbo651-byte]

🚀 The feature, motivation and pitch

If we want to generate a large dataset from dreams that can be used for training, is it possible to use an IDM model for batch inference? The current implementation neither outputs the state nor uses state information as input for inference. Can the current model meet the requirement of generating large datasets, or is it only suitable for real-time action inference scenarios?

Another question: The currently generated actions are much smaller than realistically captured actions, making the dataset completely unusable. Is there a way to resolve this?

Alternatives

No response

Additional context

No response

[Garlicnezha]

Hello, I'd like to ask how you generate actions from multi-view video input. Thank you.