… our transfer learning: 2D → 3D ConvNet performance with generic state-of-the-art 3D CNN
… weight initialization via transfer learning is possible for 3D ConvNet architecture, which can …

… fully-connected layers which perform well on transfer learning tasks [47… propose to learn
spatio-temporal features using deep 3D ConvNet. We … for video classification. We also verify that …

… replacing 3D convolutions with spatial and temporal separable 3D convolutions, … video
classification datasets Next we conduct transfer learning experiments from Kinetics to other video …

… architecture search for video classification on UCF101. We showed that our observations are
useful for spatiotemporal feature learning … be matched to a temporal slice of a 3D filter. We …

… I3D-K, we retain 3D temporal convolutions at the lowest K layers … significantly on video
classification and action detection tasks. … transfer learning experiments from Kinetics to other video …

… , we decompose a video into spatial and temporal shots. By … 3D convNet, which can recognize
human actions in videos of … size and diversity for video classification. In terms of HMDB51 …

… motion information, we propose the temporal deformable 3D ConvNets based on the attention.
The temporal deformable convolution can learn both the temporal and the spatial offsets …

… , for the video representation building based on 3D ConvNets, … spatial and temporal jittering
and different video sample rate. … to learn robust video representation for video classification. …

… Following a comprehensive study of transfer learning on ImageNet [… the-art video classification
performance in the present paper. … the spatial and temporal volume at each stacked block. …

… into the convolution network at one time, resulting in a limited temporal … 3D temporal dilation
convolution (3D-TDC) framework, to extract spatio-temporal features of actions from videos. …