dataset problem

[lwl12138997]

Hi,
thanks a lot for releasing the excellent work，I have two questions :

1. Preprocessing pipeline starting from the PerAct training dataset

In the paper you mention that you use the same training dataset as PerAct. After downloading the PerAct training dataset, did you directly run package_rlbench.py and then rearrange_rlbench_demos.py to obtain the training data format used in this repository? Or is there any additional preprocessing ?A short description of the exact preprocessing steps (e.g., which scripts to run, in what order, and what they produce) would be extremely helpful.

2. Using a trained model to generate new data in the same format

Would it be feasible to run a trained 3D Diffuser Actor policy in RLBench, record its rollouts, and then convert those rollouts into the same training dataset format that you use for training demonstrations?Do you have any recommended scripts or guidelines for turning raw rollouts (RGB / point clouds / actions / gripper states) into the same file format expected by RLBenchDataset？

Thanks again for your work and for any clarification you can provide!