Images Assisted Video Recognition by Heterogeneous Knowledge Transfer
Abstract
Intelligent systems through vision techniques aim to adaptively sense and understand the realworld scenarios. In this project, a novel heterogeneous knowledge transfer framework for images assisted video recognition is proposed. The technical merit of the proposed framework is that it can transfer knowledge between heterogeneous feature spaces. By applying different constraints on auxiliary domain, the proposed framework can be implemented in three scenarios, including domain adaption with different feature space, transfer learning with different label space, and self-taught learning with unlabeled/unrestricted image sources. These new methodologies will contribute to the robust and effective visual intelligent systems, which can deal with challenges like insufficient sampling, uncertain observations, and negative knowledge transfer.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 16, 2018
- Source ID
- W911NF1710367
Entities
People
- Yun Fu
Organizations
- Army Contracting Command
- Northeastern University
- United States Army