BriefGPT.xyz
Jun, 2024
文本图像生成模型的质量、偏差和性能分析
Analyzing Quality, Bias, and Performance in Text-to-Image Generative Models
HTML
PDF
Nila Masrourisaadat, Nazanin Sedaghatkish, Fatemeh Sarshartehrani, Edward A. Fox
TL;DR
通过 qualitatively 评估精确生成人脸、群体和指定数量对象的图片,我们不仅检查了多个文本到图像模型的性能,还进行了社会偏见分析。我们发现,容量更大的模型生成的图片质量更高,但同时这些模型也存在固有的性别或社会偏见,从而更全面地了解了它们的影响和局限性。
Abstract
Advances in
generative models
have led to significant interest in
image synthesis
, demonstrating the ability to generate high-quality images for a diverse range of text prompts. Despite this progress, most studie
→