Images without labels are considered in the training?

[LorenzoCausa]

Are images without labels considered in the training? If not, is there a way to add unlabeled images to reduce false positives?
For example, if I want to recognize rails, I don't want my model to recognize ladders as rails. To do this I would like to give a ladder image in the training data and would not label it.

[ExtReMLapin]

Yes it's ignored by default, please use cfg.DATALOADER.FILTER_EMPTY_ANNOTATIONS = True