CNN LEARNING METHOD LEARNING DEVICE FOR OPTIMIZING PARAMETERS OF CNN BY USING MULTIPLE VIDEO FRAMES AND TESTING METHOD TESTING DEVICE USING THE SAME

摘要

The present invention relates to a learning method and a learning apparatus for optimizing parameters of a CNN using a plurality of video frames, and a test method and test apparatus using the same. More specifically, (a) the CNN learning apparatus convolves the tk-th input image corresponding to the tk-th frame as a training image and the t-th input image corresponding to the t-th frame which is a frame following the tk-th frame, respectively. performing an operation at least once to obtain a tk-th feature map corresponding to the tk-th frame and a t-th feature map corresponding to the t-th frame; (b) calculating, by the CNN training apparatus, a first loss with reference to at least one distance value between each pixel of the t-k-th feature map and the t-th feature map; and (c), by the CNN training device, optimizing at least one parameter of the CNN by backpropagating the first loss. It relates to a CNN test method and test apparatus.