TNUE-Fight Detection: A New Challenge Benchmark for Fighting Recognition

Today, violence is one of the most common abnormal actions that need to be monitored and detected early. Therefore, a violence recognition system is very necessary and has great practical significance. In recent years, deep learning has achieved remarkable achievements in various different problems, however, deep learning models for violent action recognition have not been properly studied. One of the main reasons is the lack of rich labeled datasets. Specifically, the current datasets are usually only collected from movies and sports, which makes the videos in the dataset far different from actual fight actions. To overcome this drawback, in this paper, we propose a new challenge dataset named TNUE-Fight Detection. In which, our proposed dataset is collected from many real-life fights from videos uploaded to social networks. Furthermore, the TNUE-Fight Detection dataset not only provides labels for each video but also provides bounding boxes for fighting and non-fighting objects, which help to solve the problem in both cases of classification and detection. The TNUE-Fight Detection dataset is available at https://github.com/vdquang1991/TNUE_FightDetection .