GIST Youtube Anomaly Detection Dataset 2015
To evaluate the performance of the proposed method in practical scenes, we conducted experiments on the dataset which consists of videos that are downloaded from the Youtube. The dataset comprise of 3 videos for abnormal events such as a building collapses, shooting, gas explosion.
Each video includes sufficient training data to build a pattern model of normal activities. The 'building collapse(a)' video contains 805 frame, the 'shootings(b)' video consists of 3,099 frames, and the 'gas explosion (c)' video has 4,640 frames. All frames in videos was resized to the width of 320 pixels and height of 240 pixels.
[Yu et al, 2015]
Download dataset : GIST Youtube Dataset(2015).z01