Building a Video Dataset for Cricket Shot Analysis

There has been an unprecedented increase in deep learning research activity on image and video processing. However, not many video datasets are publicly available for analyzing the data. It is also observed that the quality of the available datasets is poor. Therefore, for effective learning, new training datasets have been built by the researchers, which are very often small. This effort focuses on the construction of a video dataset that can be used for cricket shot classification. An effort is made to build and benchmark a large, high-quality, balanced dataset that can be used for four types of shots. The dataset is rigorously tested with two other datasets, comprising of a dataset generated by an automated process and a benchmark dataset.