Abstract
Real-world scenarios are usually accompanied by continuously appearingclasses with scare labeled samples, which require the machine learning model toincrementally learn new classes and maintain the knowledge of base classes. Inthis Few-Shot Class-Incremental Learning (FSCIL) scenario, existing methodseither introduce extra learnable components or rely on a frozen featureextractor to mitigate catastrophic forgetting and overfitting problems.However, we find a tendency for existing methods to misclassify the samples ofnew classes into base classes, which leads to the poor performance of newclasses. In other words, the strong discriminability of base classes distractsthe classification of new classes. To figure out this intriguing phenomenon, weobserve that although the feature extractor is only trained on base classes, itcan surprisingly represent the semantic similarity between the base and unseennew classes. Building upon these analyses, we propose a simple yet effectiveTraining-frEE calibratioN (TEEN) strategy to enhance the discriminability ofnew classes by fusing the new prototypes (i.e., mean features of a class) withweighted base prototypes. In addition to standard benchmarks in FSCIL, TEENdemonstrates remarkable performance and consistent improvements over baselinemethods in the few-shot learning scenario. Code is available at:https://github.com/wangkiw/TEEN