Memory usage with batch size = 1

[zerone-fg]

Hello, I would like to ask about the approximate GPU memory usage during training when the batch size is set to 1. By the way, you mentioned in your paper that the support size during training is 64.