Is there a way to reduce false positives even when there are few negative images that can be prepared?

Depending on what you want Custom Vision to learn, you may not be able to prepare a sufficient amount of images and variations. I would like to know if there are other things that can be done in this case.