Attention driven image segmentation learning method and learning device using at least one adaptive loss weighting value map used for HD map update required to satisfy level 4 of autonomous vehicle, and testing using the same Method and testing device

摘要

At least one used to update an HD map required to meet Level 4 of an autonomous vehicle that more accurately detects blurred objects such as lanes and road signs that are visible in the distance. An attention driven image segmentation method using adaptive loss weighting values is provided. In a learning device including a CNN, in a learning process, a softmax layer is used to generate a softmax score, and a loss weight value layer is used to generate a prediction error value. Applying a calculation to generate a loss weighting value, and having a softmax loss layer, an initial softmax loss value generated by referring to the softmax score and the corresponding GT, and a loss weighting value Generating an adjusted soft max loss value with reference to. [Selection diagram] Figure 2