METHOD FOR TRAINING GENERATIVE ADVERSARIAL NETWORKS TO GENERATE PER-PIXEL ANNOTATION

摘要

An image and annotation synthesis method is disclosed. The method includes training a Generative Adversarial Network (GAN) to generate an image based on input data, acquiring an image output from the trained GAN, and acquiring an image at least one intermediate layer of the GAN. Training a decoder that outputs a semantic segmentation mask when a feature value is input based on the feature values output from and the semantic segmentation mask artificially added to the acquired image, the trained GAN and the trained GAN And generating synthesized data including at least one image and a semantic division mask corresponding to the at least one image by using a decoder.