An Image Generation Method Based on Small-Sample Continuous Learning

Abstract

The invention discloses an image generation method based on continuous learning of small samples, which includes obtaining pre-training data sets and continuous learning data sets including real images and semantic annotation maps, constructing a training system through a generative confrontation network, and guiding a sampling algorithm based on The semantic adjustment parameters of the generator are obtained from the semantic annotation map, and the training system is trained to determine the model parameters through the total loss function to obtain the image generation model. Based on the semantic annotation map of the continuous learning data set, the guided sampling algorithm is used to re-determine the generator's Semantic adjustment parameters to obtain a new semantic training system, using the total loss function to train the new semantic training system, and at the end of the training, a new semantic image generation model is obtained to complete the small-sample continuous learning of the image generation model. The method is able to continuously learn new semantics using small sample training data.

Description

technical field [0001] The invention relates to the technical field of deep learning image processing, in particular to an image generation method based on continuous learning of small samples. Background technique [0002] In recent years, Generative Adversarial Networks (GAN, Generative Adversarial Networks) have made great progress in the field of generating realistic images, which create high-quality images with rich content that humans cannot distinguish between true and false from pixel-level images. In addition, the conditional image generation method can make the generated results more controllable and meet the needs of users, for example: generating images based on text descriptions, generating human body images based on bone key points, etc. [0003] In the method of generating an image based on the semantic annotation map, each pixel in the semantic annotation map is endowed with a specific semantic meaning, so that the semantic content and layout planning of an i...

AI Technical Summary

Problems solved by technology

[0009] It can be seen that it is difficult to achieve the goals of small sample learning and continuous learning at the same time, and the simultaneous realization of the goals of small sample learning and continuous learning may affect the processing performance of the entire task sequence

Images