LEARNING METHOD AND LEARNING DEVICE FOR ATTENTION-DRIVEN IMAGE SEGMENTATION BY USING AT LEAST ONE ADAPTIVE LOSS WEIGHT MAP TO BE USED FOR UPDATING HD MAPS REQUIRED TO SATISFY LEVEL 4 OF AUTONOMOUS VEHICLES AND TESTING METHOD AND TESTING DEVICE USING THE SAME

摘要

An attention-driven image segmentation method using at least one adaptive loss weight map may be used to update the HD map required to satisfy level 4 of an autonomous vehicle. In this way, blurred objects such as lanes and road markings visible from a distance can be detected more accurately. In addition, in the military, where peer identification is important, the above method may be usefully performed to distinguish an aircraft mark or uniform from a distance. In the method, the learning device comprising: causing the softmax layer to generate a softmax score; causing a loss weight layer to generate a prediction error value, and applying a loss weight operation thereto to generate a loss weight value; and causing the softmax loss layer to generate an adjusted softmax loss value by referring to the initial softmax loss value generated by referring to the softmax score and the corresponding GT, and the loss weight value. this is provided