Video saliency target detection method based on implicit guidance and explicit teaching strategy

• Main Authors: ZHU RONG, WEI MIAN, SONG SHUQUAN, JI GEPENG, JIAO YINGXIA, NIU SHUYU
• Format: Patent
• Language: Chinese; English
• Published: 03.09.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention provides a video saliency target detection method based on implicit guidance and an explicit teaching strategy. According to the invention, a motion information modulator is introduced, attention mechanisms of two sequences of channels and spaces are included, and the motion information modulator operates in a deep cooperation mode, so that the dominant characteristics of motion information can be effectively enhanced; through an implicit guiding strategy, the characteristics of motion information leading (namely teachers) are transmitted to branches of apparent information leading (namely students), so that the detection effect can be effectively improved; in one embodiment, motion information dominant features in high three layers of a network are aggregated by using a teacher partial decoder (T-PD), and a motion information dominant mask is generated. The mask explicitly teaches the aggregation of apparent information dominant features in the high three layers of the network through another f