Bias-Awareness for Zero-Shot Learning the Seen and Unseen

Open Access
Authors
Publication date 2020
Book title 31st British Machine Vision Conference 2020
Book subtitle BMVC 2020, Virtual Event, UK, September 7-10, 2020
Event 31st British Machine Vision Conference
Article number 261
Number of pages 13
Publisher BMVA Press
Organisations
  • Faculty of Science (FNWI) - Informatics Institute (IVI)
Abstract
Generalized zero-shot learning recognizes inputs from both seen and unseen classes. Yet, existing methods tend to be biased towards the classes seen during training. In this paper, we strive to mitigate this bias. We propose a bias-aware learner to map inputs to a semantic embedding space for generalized zero-shot learning. During training, the model learns to regress to real-valued class prototypes in the embedding space with temperature scaling, while a margin-based bidirectional entropy term regularizes seen and unseen probabilities. Relying on a real-valued semantic embedding space provides a versatile approach, as the model can operate on different types of semantic information for both seen and unseen classes. Experiments are carried out on four benchmarks for generalized zero-shot learning and demonstrate the benefits of the proposed bias-aware classifier, both as a stand-alone method or in combination with generated features.
Document type Conference contribution
Language English
Other links https://github.com/twuilliam/bias-gzsl https://dblp.org/db/conf/bmvc/bmvc2020.html https://www.bmvc2020-conference.com/programme/accepted-papers/
Downloads
0261 (Final published version)
Permalink to this page
Back