Class Anchor Clustering: a Loss for Distance-based Open Set Recognition
This addresses the challenge of distinguishing unknown classes in deep neural networks for computer vision, representing a novel method rather than an incremental improvement.
The paper tackles the problem of open set recognition by introducing the Class Anchor Clustering (CAC) loss, which explicitly trains known classes to form tight clusters in logit space, achieving state-of-the-art performance with a 15.2% AUROC increase on TinyImageNet without compromising classification accuracy.
In open set recognition, deep neural networks encounter object classes that were unknown during training. Existing open set classifiers distinguish between known and unknown classes by measuring distance in a network's logit space, assuming that known classes cluster closer to the training data than unknown classes. However, this approach is applied post-hoc to networks trained with cross-entropy loss, which does not guarantee this clustering behaviour. To overcome this limitation, we introduce the Class Anchor Clustering (CAC) loss. CAC is a distance-based loss that explicitly trains known classes to form tight clusters around anchored class-dependent centres in the logit space. We show that training with CAC achieves state-of-the-art performance for distance-based open set classifiers on all six standard benchmark datasets, with a 15.2% AUROC increase on the challenging TinyImageNet, without sacrificing classification accuracy. We also show that our anchored class centres achieve higher open set performance than learnt class centres, particularly on object-based datasets and large numbers of training classes.