LEARNING METHOD AND LEARNING DEVICE FOR IMPROVING SEGMENTATION PERFORMANCE TO BE USED FOR DETECTING EVENTS INCLUDING PEDESTRIAN EVENT VEHICLE EVENT FALLING EVENT AND FALLEN EVENT USING EDGE LOSS AND TEST METHOD AND TEST DEVICE USING THE SAME

摘要

A learning method for improving segmentation performance, which is used to detect events such as pedestrian events, car events, polling events, and pollen events, using a learning device, is provided. The method includes the steps of: (a) causing k convolutional layers to generate k encoded feature maps; (b) (k-1) deconvolution layers sequentially generate (k-1) decoded feature maps, and the learning device causes h mask layers to be generated from h deconvolution layers corresponding thereto. Referring to h edge feature maps generated by extracting edge portions from the output h basic decoded feature maps and the h basic decoded feature maps; And (c) causing the h edge loss layers to generate h edge loss with reference to the edge portion and the corresponding GT. In addition, the method can increase the degree of detection of traffic signs, landmarks and road signs.