Graph Feature Generation based on Scene Graph Benchmark Application on Video

Abstract

In the field of vision research, it is considered that the analysis of scene graphs can help machines comprehend higher-level visual scene contexts. The data structure that both detects objects and defines their relationships is called a scene graph. The scene graph has been applied to diverse tasks, including action recognition, image captioning and visual question answering. It is also suggested as a method for understanding the video context. The amount of video data needed for training to produce a scene graph is nonetheless insufficient. Therefore, the inference model was applied to real-world data, and the possibility of usage of scene graph as a visual feature was identified.