Sports Video Summarization with Limited Labeling Datasets Based on 3D Neural Networks

As more and more information generated, recorded, and distributed daily, the video summarization allows the spectators to focus on the clips mostly containing the important events for watching time saving without scarifying the video quality. The purpose of this work is to use the deep learning for removing the redundant video clips, whereas keeping the critical content exhaustive. We would try to bridge the semantic gap between the low-level video features and the high-level human perception, which is prone to be modeled by the neural network systems. The advantages will also be verified by comprehensive experiments for MLB (Major League Baseball) video summarization in terms of reduced size and expressional consistency.