Activities in Extended Video.
Fan Ma, Peike Patrick Li, Linchao Zhu, Xuanyi Dong, Yanbin Liu, Yi Ping Yang · UTS ePRESS (University of Technology Sydney) · 2018
In this paper, we present a system based on detection, tracking and 3D convolution neural network dealing with Activities in Extended Video (ActEV) task in TRECVID 2018. In the proposed system, videos are first unfolded into frames for training detection network, then we use it to generate bounding box for tracking areas where target activities could be happen. The tracking clips are then classified using a 3D convulution network.