A learning method and a learning device for converting a real image in real world into a virtual image in a virtual world using a cycle Gan which can be applied to a domain adaptation to detect an object on a real image Test method and test device utilizing the same

摘要

To provide a learning method, a learning device, a test method, and a test device which can reduce difference between virtual and reality and reduce annotation costs, and which learns runtime input transformation for transforming real images into virtual images.SOLUTION: A learning method includes steps of: allowing a first transformer to transform a first image to a second image, allowing a first discriminator to generate a 1_1-st result, and allowing a second transformer to transform the second image to a third image, having characteristics which are same as or similar to real images; allowing the second transformer to transform a fourth image to a fifth image, allowing a second discriminator to generate a 2_1-st result, and allowing the first transformer to transform the fifth image to a sixth image; and calculating losses.SELECTED DRAWING: Figure 3