Improving Video Retrieval with Data Expansion and Knowledge Graphs
Özet
Recent video-text multimodel approaches provide good results on retrieval tasks. However, the performance of domain-specific cases might be decreased because of internal differences with generic data. Since deep learning models are data-hungry methods, automatic data augmentation can be a solution. Domain-specific data has internal structured information and can be used to increase the retrieval performance. Hence another option is using this structured information to boost the performance of the retrieval. In this work, we show an automatic data expansion pipeline for cooking videos. Then we investigate how utilizing domain-specific knowledge graphs can be a solution.
