Table 2.

Summary of baseline methods.

Method	Core Idea	Open-Set Discrimination Mechanism
SoftMax-T [23]	Standard cross-entropy classification.	Compares the maximum SoftMax probability against an empirical threshold.
OpenMax [24]	Calibrates the activation values of the SoftMax output layer.	Utilizes extreme value theory (EVT) to model the tail of activation scores for each class, calculating the probability of unknown classes.
MLOSR [28]	Combines classification and reconstruction tasks.	Uses extreme value theory (EVT) to model the tail of reconstruction errors for all known classes.
SR2CNN [33]	Multi-task learning combining classification, reconstruction, and center loss.	Compares the minimum distance between test samples and class centers against an empirical threshold.
TripletNet [35]	Metric learning based on triplet loss.	Compares the average distance between test samples and their k-nearest neighbors against an empirical threshold.