Please follow all rules for the FGVC11 Workshop.

In addition, you should train your model only on the provided training set containing scans of individual bugs. The validation set with the scans of bug mixtures is provided only for validating your predictions and making sure that you comply with the submission template. Therefore, you are **not** allowed to train on the validation data of mixtures. The BugNIST2024 competition evaluates domain shift, and this will not be the case if you use the validation data while training.

You are free to make synthetic mixtures using the provided training set of individually scanned bugs. This can be a useful data augmentation for training your model.