Multi-Level Feature Fusion Using Hierarchical Graph Network for Video Summarization

The exponential increase in video-based content has increased the demand for sophisticated video summarization techniques for efficient browsing and retrieval. These techniques aim to compress extended videos into brief summaries by intelligently selecting representative segments that encapsulate the essence of the original content while retaining its vital information and narrative structure. In this work, a hierarchical graph neural network-based model is proposed which aims to extract both fine-grained and high-level features in a video by capturing and combining spatial and temporal information at multiple levels by modeling both intra-segment and inter-segment interactions. The model's performance is evaluated on two standard benchmark datasets, SumMe and TVSum, demonstrating its performance in generating high-quality summaries and attaining competitive accuracy.