Unsupervised Domain Adaptation for Training Event-Based Networks Using Contrastive Learning and Uncorrelated Conditioning
This work addresses a domain-specific challenge for event-based vision, enabling deep learning applications in high-dynamic range environments, though it appears incremental as it builds on existing domain adaptation techniques.
The paper tackles the problem of annotated data scarcity for event-based cameras by developing an unsupervised domain adaptation algorithm using contrastive learning and uncorrelated conditioning, which outperforms existing methods for event-based image classification.
Event-based cameras offer reliable measurements for preforming computer vision tasks in high-dynamic range environments and during fast motion maneuvers. However, adopting deep learning in event-based vision faces the challenge of annotated data scarcity due to recency of event cameras. Transferring the knowledge that can be obtained from conventional camera annotated data offers a practical solution to this challenge. We develop an unsupervised domain adaptation algorithm for training a deep network for event-based data image classification using contrastive learning and uncorrelated conditioning of data. Our solution outperforms the existing algorithms for this purpose.