GAN LEARNING METHOD AND LEARNING DEVICE FOR GENERATION OF VIRTUAL FEATURE MAPS WHOSE CHARACTERISTICS ARE SAME AS OR SIMILAR TO THOSE OF REAL FEATURE MAPS BY USING GAN CAPABLE OF BEING APPLIED TO DOMAIN ADAPTATION TO BE USED IN VIRTUAL DRIVING ENVIRONMENTS

摘要

The present invention uses a Generative Adversarial Network (GAN) including a Generating Network and a Discriminating Network, which can be applied to domain adaptation used in a virtual driving environment, to generate a real image (Real It relates to a learning method for deriving a virtual feature map from a virtual image having the same or similar characteristics as a real feature map derived from the image, (a) Causing the learning device to cause the generating network to apply a convolution operation to the input image, thereby generating an output feature map having characteristics identical or similar to those of the actual feature map; And (b) causing the first loss unit to generate a loss with reference to an evaluation score generated by the delimiting network, corresponding to the output feature map. Characterized in that, through the above method used for runtime input conversion, it is possible to reduce the difference between the virtual and the reality and the annotation (Annotation) cost.