Exemplar-Free Deep Incremental Hashing for Efficient Image Retrieval
摘要
Deep hashing techniques have been advanced by CNNs’ semantic representations. However, existing incremental hashing methods rely on original data to maintain similarities, which is often inaccessible due to privacy, legal, and transmission constraints. To address this, we introduce EFIH, an exemplar-free deep incremental hashing approach. EFIH learns hash codes for new classes without old samples, using knowledge distillation to mimic old models and a prototype-based loss to preserve similarities. For unavailable data, we adopt self-supervised label augmentation. Experiments show EFIH’s superiority and effectiveness.