ACGAN论⽂笔记
《》
Odena A, Olah C, Shlens J. Conditional Image Synthesis With Auxiliary Classifier GANs[J]. 2016.
⽬的:合成(synthesis)⾼分辨率图像
⽅法:提出⼀种新的变种条件标签GAN
贡献:提出ACGAN,提出新的⽣成图⽚质量(可判别性和多样性 discriminability and diversity of samples)评估⽅法Inception Accuracy和MS-SSIM,得出结论:⾼分辨图⽚在ACGAN下可以更为准确地做classification(Across 1000 ImageNet classes, 128 _ 128 samples are more than twice as discriminable as artificially resized 32 _ 32 samples. In addition, 84.7% of the classes have samples exhibiting diversity comparable to real ImageNet data.1)。
核⼼:In this work we demonstrate that that adding more structure to the GAN latent space along with a specialized cost function results in higher quality samples.
模型架构:
resized
Loss函数:
The objective function has two parts: the log-likelihood of the correct source, LS, and the log-likelihood of the correct class, LC.
D is trained to maximize LS + LC while G is trained to maximize LC -LS.
Inception accuracy:
OpenAI提出Inception Score⽤以评价合成图像的真假,本⽂作者认为在⽣成图像不符合⼈的视觉感知时,Inception Score依然可能会很⾼,所以评判不准确。于是,作者将合成的图像输⼊训练好的Inception V3模型,对⽐模型的分类准确率。
不同分辨率的⽣成图⽚的Inception accuracy对⽐,⼈⼯降低分辨率后,accuracy也随之降低,说明⽣成的⾼分辨率图⽚不仅含有像素信息,还有⼀些帮助提⾼accuracy的信息。
总结:ACGan⽴⾜添加标签约束,以提⾼(⾼分辨率)图⽚⽣成质量,并提出新的⽣成图⽚质量和模式坍塌衡量标准。作者在附录中阐明了标签数量对模型稳定性的影响,实验中训练了100个AC-GAN,每个对10个标签进⾏分类。本⽂没有过多与其他现有的⽣成模型进⾏⽐较测试。

版权声明:本站内容均来自互联网,仅供演示用,请勿用于商业和其他非法用途。如果侵犯了您的权益请与我们联系QQ:729038198,我们将在24小时内删除。