Summarizing and Searching Video: Domain Adaptation
October 2019 • Poster
This poster describes research to enable the use of existing data sets to training detectors in low data environments.
Software Engineering Institute
Despite impressive improvements in machine learning systems in recent years, classifiers still struggle to perform when there is little or no training data in the target environment. Semantic differences, such as perspective and object density, between source and target environments can significantly degrade classifier accuracy. Non-semantic differences, such as differences in object environment, can significantly degrade classifier accuracy. Differences between the trained and real-world data sets also hamper classifier performance.