知识卡片
生成器摊销逐样本优化
传统统计纹理合成可为每张输出从噪声出发单独优化像素;GAN则训练一个共享映射,之后用前向运行生成不同样本,把许多次构造的代价摊到训练中。共享快速生成带来效率,也可能受训练覆盖和映射容量限制。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 生成模型
原书出处 · 1 处
- · 查看原书依据
GANs are related to the image and texture models we saw in chapters 27 and 28. For example, in the Heeger-Bergen model (section 28.3; [202]), we synthesize images with the same statistics as a source texture. This can be phrased as an optimization problem in which we optimize image pixels until certain statistics of the images match those same statistics measured on a source (training)
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第32章《生成模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c032.xhtml)
- 结论依据:摘录中关于生成器摊销逐样本优化的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:GANs are related to the image and texture models we saw in chapters 27 and 28. For example, in the Heeger-Bergen model (section 28.3; [202]), we synthesize images with the same statistics as a source texture. This can be phrased as an optimization problem in which we optimize image pixels until certain statistics of the images match those same statistics measured on a source (training)