Large-scale Video Classification with Convolutional Neural Networks, CVPR 2014

To Appear at CVPR 2014. More information on http://cs.stanford.edu/people/karpathy/deepvideo Note that the temporal smoothing is applied in a 200-Frame (~6 second) gaussian window centered on the frame that is shown. In other words, the network is making its decision at time t based on frames [t-100...t+100].