Bmt-Bench: A Benchmark Sports Dataset For Video Generation

In recent years, there has been a growing interest among researchers and scholars in the analysis of sports activities, driven by the advancements of machine learning and the increased availability of public data. However, there remains a scarcity of comprehensive sports video datasets that possess the necessary attributes to address various research tasks effectively. We present the “Badminton Benchmark” (BMT-BENCH) to facilitate reproducible machine learning research in the sports domain. This dataset comprises high-quality, high-speed video clips collected from official badminton tournaments involving two team players. The dataset is labeled and unlabeled, catering to different research problems such as video generation and real-time object detection. we feature a baseline system mainly for video generation tasks and provide a thorough evaluation of the challenges posed by the dataset’s unique nature. The dataset is publicly accessible at https://drive.google.com/drive/folders/1moYDb8tp5K-VDxPJU3sTorfY E7NnwVpf?usp=sharing and the baseline system is available at https://github.com/ziangshi/BMT_BENCH_baseline_repo.