How to deal with multiple dataloaders?

vision
1

Hi.

I am trying to train a multiple dataloader using multiprocessing.

I am trying to train by creating multiple dataloaders set to the input_size and batch_size I want.

I have two questions.

  1. If it is assumed that 3 dataloaders are created, is memory allocated to each dataloader immediately after creation?

  2. Are there examples of controlling multiple dataloaders using multiprocessing?

Thank you in advance.

2

For 1: Have a look at this post, which explains the memory usage depending on the implementation of the Dataset.

2: I don’t know how multiprocessing would work on top of the DataLoader, which itself uses multiprocessing for the workers.