SPC-FastGAN: Semantic-Projected Consistency FastGAN for Fine-Grained Few-Shot Image Generation
Zejia Zhang, Mingyang Li, Liming Wang, Jiehui He, Huirong Hu
Guangdong University Of Finances and Economics Changchun University of Technology
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
To address the problem that lightweight FastGAN suffers from overfitting and training instability when performing few-shot fine-grained image generation, as its discriminator tends to memorize local textures verbatim, this paper proposes a Semantic Projection Consistency improved framework (SPC-FastGAN).This framework non-intrusively preserves the original lightweight backbone. On one hand, it extracts stable high-level semantic information by freezing the pre-trained semantic projection branch; on the other hand, it introduces a projection space consistency regularization constraint to precisely suppress discriminator overfitting.Experiments on 10/50/100-shot settings demonstrate that the proposed method significantly optimizes the FID and KID metrics, and exhibits exceptional resistance to training collapse especially in the 10-shot extremely low-shot scenario.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIGenerative Adversarial Networks and Image Synthesis
Domain Adaptation and Few-Shot Learning · Advanced Neural Network Applications