知识卡片

普通读书笔记卡 · 8980.a · 生成模型:GPT-6

潜在坐标不保证语义解耦

理解等级:中级

生成器的潜变量控制输出变化,但某一维不必独自代表颜色、姿态或背景,不同因素也可能纠缠在多维中。固定z便于重现同一确定生成过程,仍需同时固定模型及其他随机来源。语义编辑能力需要额外结构或验证。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Usually we draw z from a simple distribution such as Gaussian noise, that is, p(Z) = N (0, 1). The way to think of z is that it is a vector of latent variables, which specify all the attributes of the images other than those determined by the label. These variables are called latent because they are not directly observed in the training data (it

    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:摘录中关于潜在坐标不保证语义解耦的定义、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:Usually we draw z from a simple distribution such as Gaussian noise, that is, p(Z) = N (0, 1). The way to think of z is that it is a vector of latent variables, which specify all the attributes of the images other than those determined by the label. These variables are called latent because they are not directly observed in the training data (it