[ECCV 2020] Class-Incremental Domain Adaptation (10 min talk)

ECCV 2020

We introduce Class-Incremental Domain Adaptation (CIDA) that enables the recognition of both shared and novel target categories under a domain-shift. Project page: https://sites.google.com/view/cida-eccv arXiv: https://arxiv.org/pdf/2008.01389.pdf Authors: Jogendra Nath Kundu, Rahul Mysore Venkatesh, Naveen Venkat, Ambareesh Revanur, R. Venkatesh Babu Indian Institute of Science, Bengaluru Abstract: 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. Citation: @inproceedings{kundu2020class, title={Class-Incremental Domain Adaptation}, author={Kundu, Jogendra Nath and Mysore Venkatesh, Rahul and Venkat, Naveen and Revanur, Ambareesh and Babu, R. Venkatesh}, inproceedings={European Conference on Computer Vision (ECCV)}, year={2020} }