Cricket Actions Dataset

Cricket Actions Dataset 是由信息与技术朗姆米特学院创建的一个大规模未修剪的板球视频数据集，包含约7300万帧和1110个视频。该数据集主要用于板球动作的时序定位研究，特别是板球击球动作的提取。数据集通过一系列预处理步骤，包括统一帧率和分辨率，以确保数据的一致性。创建此数据集的目的是为了训练深度神经网络，解决自动内容基础的板球击球动作识别问题，进而应用于自动解说生成和视频内容浏览与索引。

Cricket Actions Dataset is a large-scale untrimmed cricket video dataset created by Rumjimet Institute of Information and Technology, which contains approximately 73 million frames and 1110 video clips. This dataset is primarily intended for temporal localization research on cricket actions, particularly the extraction of cricket batting motions. A series of preprocessing steps, including unifying frame rates and resolutions, are implemented to ensure data consistency. The dataset is developed to train deep neural networks for solving the problem of automatic content-based cricket batting action recognition, and can be further applied to automatic commentary generation, as well as video content browsing and indexing.