STRAN: Student expression recognition based on spatio-temporal residual attention network in classroom teaching videos

Improved C3D Network Model Construction and Its Posture Recognition Study in Swimming Sports

“…Among them, the DL algorithm has become a commonly used method in the research of S-Pos recognition. The C3D network in DL can realize the direct extraction of ST features, which effectively circumvents the defects of the dual-stream network that consumes a large number of resources in order to realize the extraction of the temporal features individually [16][17]. However, the huge number of parameters can cause the convolutional network to be difficult to extract ST features completely during the extraction process, and the effectiveness of feature extraction is limited by the narrow number of convolutional network layers.…”