Finetuning hyperparameter w about ddad HOT 3 CLOSED

Hello, appreciate your inquiry.

Setting w is based on hyperparameters optimisation. We have not proposed a solid solution to set it automatically and it can be an interesting avenue for future works.

I hope this explanation adequately addresses your question.

from ddad.

Thank you for your reply, what data do you use for the optimization of w?

from ddad.

If no validation set is available for hyperparameter tuning, a prudent option is to employ synthetic anomalies. Some papers, such as DRAEM, utilize DTD as synthetic data during training. The same synthetic anomalies can be adopted to identify the optimal configuration in the absence of a validation set.

DRAEM: https://arxiv.org/abs/2108.07610
DTD: https://www.robots.ox.ac.uk/~vgg/data/dtd/

from ddad.