search

liznerski / fcdd

Repository for the Explainable Deep One-Class Classification paper
MIT License
225 stars 62 forks source link

noise_mode for custom data #67

Closed NimaDL closed 1 year ago

NimaDL commented 1 year ago

Hi

I was wondering which noise mode should I choose to better fit the model on my custom data (the data is from completely different domain than ImageNet). I want to train the model using semi-supervise approach where I show only few anomalous data to the model during the training phase. I am a bit confused what noise model should be chosen? In this case isn't better not to use noise mode option? I am not sure if this option is possible in this version of FCDD!?

liznerski commented 1 year ago

Use --supervise-mode other. In that case, --noise-mode is ignored since the training will use ground-truth anomalies. Follow the description provided here on where to place the ground-truth anomalies for training.