An attention driven image segmentation learning method and a learning device utilizing at least one adaptive loss weighted value map utilized for HD map update required to meet the level 4 of an autonomous vehicle Testing method and testing device using the same

摘要

To provide an attention driven image segmentation method using at least one adaptive loss weighting value which is used for updating an HD map required for satisfying a level 4 of an autonomous travel vehicle for more correctly detecting a dim object such as a lane or a road sign visible in the distance.SOLUTION: A learning device including a CNN is configured so that, a learning process comprises: a step for using a soft max layer for generating a soft max score; a step for using a loss weight layer for generating a prediction error value, and applying a loss weighting value calculation to the prediction error value for generating a loss weighting value; and a step for using a soft max loss layer, for referring to an initial soft max loss value generated by referring to a soft max score and a GT corresponding to the same, and the loss weighting value, for generating an adjustment soft max loss value.SELECTED DRAWING: Figure 2