知识卡片

普通读书笔记卡 · 9098.a · 生成模型:GPT-6

VAE与扩散采样路径不同

理解等级:中级

基本VAE可由潜变量和条件经一次解码得到图像,扩散则通过多步去噪逐渐形成样本。二者都可从噪声和文本出发,但计算路径、训练目标与细节表达不同;不能仅凭这一共性认为实现或结果质量相同。

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

      Notice that figure 51.10 is actually very similar to figure 51.9. A conditional diffusion model is in fact very similar to a cVAE. At sampling time, both take as input text and Gaussian noise. The VAE uses a network that directly maps these inputs to an image, in a single forward pass. The diffusion model, in contrast, maps to an image via a chain of

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第51章《视觉与语言》(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/c051.xhtml) - 结论依据:摘录中关于VAE与扩散采样路径不同的定义、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:Notice that figure 51.10 is actually very similar to figure 51.9. A conditional diffusion model is in fact very similar to a cVAE. At sampling time, both take as input text and Gaussian noise. The VAE uses a network that directly maps these inputs to an image, in a single forward pass. The diffusion model, in contrast, maps to an image via a chain of