Learning method and apparatus for allowing CNN learned in virtual world to be used in real world by conversion of runtime input using photo style conversion, and testing method and testing apparatus using the same

摘要

A CNN learned in a virtual world is used in the real world. A learning device generates a first predicted autonomous driving source information by referring to a step of acquiring a first learning image of virtual traveling of a virtual vehicle and the first learning image, and a first original corresponding to the first prediction. After the first main loss is generated by referring to the correct answer autonomous driving source information, the first learning process is performed by performing the back propagation using the first main loss and the first learning process and the support CNN correspond to the first learning image. The second original image correct autonomous driving source corresponding to the second prediction after generating the second learning image by referring to the image of the first base and the second base of the real driving of the actual vehicle and generating the second predicted autonomous driving source information After generating the second main loss with reference to the information, performing a second learning process of learning parameters by performing back propagation using the second main loss. [Selection diagram] Figure 2