Class-Incremental Domain Adaptation
Jogendra Nath Kundu* Rahul Mysore Venkatesh* Naveen Venkat Ambareesh Revanur R. Venkatesh Babu
Video Analytics Lab, Indian Institute of Science, Bangalore
Accepted in ECCV 2020
We introduce a practical Domain Adaptation (DA) paradigm called Class-Incremental Domain Adaptation (CIDA). Existing DA methods tackle domain-shift but are unsuitable for learning novel target-domain classes. Meanwhile, class-incremental (CI) methods enable learning of new classes in absence of source training data, but fail under a domain-shift without labeled supervision. In this work, we effectively identify the limitations of these approaches in the CIDA paradigm. Motivated by theoretical and empirical observations, we propose an effective method, inspired by prototypical networks, that enables classification of target samples into both shared and novel (one-shot) target classes, even under a domain-shift. Our approach yields superior performance as compared to both DA and CI methods in the CIDA paradigm.
Problem Setting. A) Closed-set DA assumes a shared label-set between the source and the target domains. B) Open-set DA rejects target samples from unseen categories into a single unknown class. C) In Class-Incremental DA (CIDA), we aim to recognize both shared and new target classes by assigning a unique semantic label to each class.