Question regarding downsampling and resizing dataset

Hi,

Thank you for providing this wonderful library! I was curious about how one can take an existing dataset (collected or downloaded) and modify the fps (downsample, resize images, or delete specific episodes (for v3) prior to policy training. I am finding this tricky to do particularly when the dataset is not loaded in code but provided as a parameter to lerobot-train. I've spent time digging around the codebase but didn't see a way that doesn't involve loading the dataset in script first and adjusting this (for resizing, not sure about downsampling fps). Does the codebase provide utility functions for this? Thanks!

Start by tracing the lerobot-train dataset parameter and the existing dataset-loading path. Check how resizing, FPS downsampling, and episode deletion are currently handled, including the user's observation that resizing requires a script. Done should be a clear supported workflow or a scoped utility proposal for preprocessing datasets before policy training.

Written by the indexing model from the issue text.