The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) recently released the paper acceptance result. A study on video understanding carried out by Associate Professor Zhang Yifeng from the Artificial Intelligence Research Center of the School of Information Science and Engineering was enrolled by the conference. Liu Yuan, a master's student specializing in signal processing isthe first author and his instructor Zhang Yifeng is the co-correspondent.

Liu Yuan’s paper “Multi-granularity generator for temporal action proposal” is a research on video understanding so as to find the specific time segments of human motion in long videos. Currently, the motion segments cannot be positioned accurately and it is impossible to find the time segments of different lengths, etc.. This study was carried out on the basis of multi-granularity analysis of the entire video, in which the coarse granularity generator is used to find the motion time segments of different lengths, and the fine granularity generator is more capable of accurate time positioning. The combination of the two can position the start and end time of the motion accurately. Figure 1 presents the research ideas of the paper and is also the core innovation of the study. Figure 2 shows the overall structure of the research network. This research will play a fundamental role in scene identification and dangerous action warning in the security field and also play an important role in the construction of “smart cities”.

It is reported that CVPR is the world's top conference on computer vision and pattern recognition. It guides the future research direction of computer vision. It is listed as a Class A conference by the Chinese Computer Society (CCF) with the annual paper acceptance rate of generally around 20%. According to the currently prevailing Google Scholar Citation statistics, CVPR’s H5 index is 158, ranking 35th in all subject catalogues (Nature first, Science third, and Cell sixth) and the top among all journals and conferences in the telecommunications discipline. This research achievement marks the continuous innovation and progress of the School of Information Science and Engineering in the field of computer vision and artificial intelligence. CVPR-2019 International Conference will be held in Long Beach, USA from June 16 to June 20, 2019.