Related Experiment Videos
Video summarization based on multi-scale feature fusion
Jing Bao1, Shipeng Xu1, Jing Zhang1
1School of Cyberspace Security, Gansu University of Political Science and Law, Lanzhou, Gansu, China.
Plos One
|July 23, 2026
Summary
This study introduces a novel multi-scale feature fusion model for video summarization. It effectively balances local and global features, improving concise video representation and content extraction.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Existing graph-based video summarization methods struggle with multi-scale feature interactions and balancing local/global features.
- Efficient video summarization is crucial for quick information retrieval and content understanding.
Purpose of the Study:
- To propose a multi-scale feature fusion video summarization model using the Message Passing Neural Network (MPNN) framework.
- To address limitations in current methods by enhancing the modeling of feature interactions at various scales.
Main Methods:
- Extracted representative frame features from video shots, creating a graph with shot features as nodes and semantic similarity as edges.
- Employed MPNN to decouple multi-scale features, using Graph Attention Network (GAT) for local features and Graph Neural Network (GNN) for global features.
- Fused original, local, and global features to calculate shot importance and generate summaries.
Main Results:
- The proposed Multi-Scale Feature Fusion-Message Passing Neural Network (MSF-MPNN) model achieved competitive performance on SumMe (F1=50.0) and TVSum (F1=61.8) datasets.
- Demonstrated superior performance compared to mainstream Recurrent Neural Network (RNN)/GNN-based methods.
- Successfully balanced local correlation and global features for improved summarization.
Conclusions:
- MSF-MPNN offers a practical and effective solution for efficient video content extraction.
- The model shows significant potential for applications in video surveillance and short-video platforms.
- Highlights the effectiveness of multi-scale feature fusion in graph-based video summarization.