A learning method and apparatus for a runtime input transformation for transforming a real image in a real world into a virtual image in a virtual world, which is used for object detection on a real image using a cycle GAN applicable to domain adaptation, and Test method and test equipment using it

摘要

PROBLEM TO BE SOLVED: To provide a learning method and a learning device, a testing method and a testing device for learning a runtime input conversion for converting a real image into a virtual image, which can reduce a difference between virtual and real and annotation cost. According to a learning method, a first transformer (Transformer) transforms a first image into a second image, a first discriminator generates a first_1 result, and a second transformer transforms a second image. Converting a fourth image into a fifth image with a second converter, and generating a second_1 result with a second discriminator, Transforming the fifth image into a sixth image with one converter and calculating the loss. [Selection diagram]