BriefGPT.xyz
Jun, 2020
超分辨率变分自编码器
Super-resolution Variational Auto-Encoders
HTML
PDF
Ioannis Gatopoulos, Maarten Stol, Jakub M. Tomczak
TL;DR
通过在变分自编码器中添加缩小版本的图像随机变量，提高图像生成的质量表现，使该方法在负对数似然方面和VAE表现相当，同时在数据合成方面获得更好的FID得分。
Abstract
The framework of
variational autoencoders
(VAEs) provides a principled method for jointly learning
latent-variable models
and corresponding inference models. However, the main drawback of this approach is the blu
→