Scholarlink

This paper aims to research and develop methods for pre-training generative adversarial models for image translation tasks in data-limited environments. In particular, the research aims to develop strategies for training data augmentation to improve the efficiency of training such models in situations where the number of available translated images is limited. To achieve this goal, we analyze the CycleGAN model's architecture and available methods for pre-training generative adversarial models for image translation and identify their limitations and advantages. New data augmentation methods are developed and evaluated to improve the training results of generative models with limited training images.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Pre-training Generative Adversarial Models for Image Translation Under Limited Data Conditions

  • 章节

  • 发表日期：

    2026年

查看全文
  • Yurii Kryvenchuk

摘要

This paper aims to research and develop methods for pre-training generative adversarial models for image translation tasks in data-limited environments. In particular, the research aims to develop strategies for training data augmentation to improve the efficiency of training such models in situations where the number of available translated images is limited. To achieve this goal, we analyze the CycleGAN model's architecture and available methods for pre-training generative adversarial models for image translation and identify their limitations and advantages. New data augmentation methods are developed and evaluated to improve the training results of generative models with limited training images.