LEARNING METHOD AND LEARNING DEVICE FOR RUNTIME INPUT TRANSFORMATION OF REAL IMAGE ON REAL WORLD INTO VIRTUAL IMAGE ON VIRTUAL WORLD, TO BE USED FOR OBJECT DETECTION ON REAL IMAGES, BY USING CYCLE GAN CAPABLE OF BEING APPLIED TO DOMAIN ADAPTATION

摘要

The present invention relates to a method for learning a runtime input transformation to transform a real image into a virtual image using a cycle GAN that can be applied to domain adaptation, which can be performed in a virtual driving environment, (a) (i) 1 cause a Transformer to transform the first image into a second image, (ii) (ii-1) cause a first discriminator to generate a first_1 result, (ii-2) causing the second converter to convert the second image into a third image having the same or similar characteristics to the real image; (b) (i) cause the second transformer to transform the fourth image into the fifth image, (ii) (ii-1) cause the second discriminator to produce a result 2_1, (ii-2) causing the first converter to transform the fifth image into a sixth image; and (c) calculating the loss; this method can reduce the difference between virtual and reality and the cost of annotation.