Key points are not available for this paper at this time.
Action recognition task involves the identification of different actions from video clips where the action mayor may not be performed throughout the entire duration of the video. Despite the stratospheric success of deep learning architectures in image classification (ImageNet), progress in architectures for video classification and representation learning has been slower. This paper proposes a new usage of the Two-Stream Inflated 3D ConvNet (I3D) that is based on 2D ConvNet inflation to detect action in multi-action videos. Validation results are very promising where it shows accuracy up to 93%. One of the most important applications of this work is security in smart homes.
Rami et al. (Thu,) studied this question.