Different kitchen datasets for different methods

[ziyadsheeba]

I was lookin through the Franka Kitchen experiments, and realized that in your training logs here, you use the dataset diffusion_policy.dataset.kitchen_lowdim_dataset.KitchenLowdimDataset for all the other methods but for diffusion policy you are using class diffusion_policy.dataset.kitchen_lowdim_dataset.KitchenMjlLowdimDataset. I checked the parsed mjl files and there are 605 trajectories compared to the 566 trajectories in the KitchenLowdimDataset. Why are you not using the same dataset for all the different methods? It seems like having 40 more trajectories does not result in a fair comparison.

Thanks!