【摘要】 技术能降低原始视频冗余程度，减小视频存储空间和浏览时间。现有的视频摘要方法大多采用递归结构，不仅计算复杂而且模型很难并行化运行。为了解决这个问题，提出了一种基于门控多头注意力机制的视频摘要网络模型（GMPAVS），该模型包含两种多头自注意模块，能同时捕获时间帧之间的全局和局部依赖关系，同时引入了相对位置编码和绝对位置编码，用于捕获视频的时间相关性。在两个基准数据集SumMe和TVSum上的实验结果证明了该方法的有效性。更多还原

【Abstract】 Video summarization technology can reduce the redundancy of original videos as well as video storage space and browsing time. Most of the existing video summarization methods adopt a recursive structure, which is not only computationally complex but also makes it difficult to run the model in parallel. To address this issue, this paper proposes a video summarization network model(GMPAVS) based on a gated multi-head attention mechanism. The model contains two multihead self-attention modules, which can simultaneously capture the global and local dependencies between time frames, and introduce relative position encoding and absolute position encoding to capture the temporal correlation of videos.更多还原