Is the entire dataset open-sourced?
The paper mentions three parts of data, including trajectory and non-trajectory data. However, it seems that only the trajectory data is available in the open-source dataset? Also, the training shell script seems to freeze the visual encoder, corresponding only to the final training stage mentioned in the paper?
Is the entire dataset open-sourced?
The paper mentions three parts of data, including trajectory and non-trajectory data. However, it seems that only the trajectory data is available in the open-source dataset? Also, the training shell script seems to freeze the visual encoder, corresponding only to the final training stage mentioned in the paper?