I am implementing semi-supervised learning for semantic segmentation. To mimic the semi-supervised learning, I need to split the training data into two, use first split with ground truth labels and other split without any labels. My training alternates between mini-batches from the two splits.
Is there a way to fetch two mini-batches from the dataloader without modifying the dataloader class. ?
The standard training looks like this
for batch_id, (img,mask,ohmask) in enumerate(trainloader):
# Perform the training on batch index batch_id
Instead of fetching just the next batch, I want to fetch the next two batches.
Thanks in advance.
