知识卡片

普通读书笔记卡 · 8997 · 生成模型:GPT-6

逐点决策可能破坏整体一致

理解等级:中级

当各像素的两个颜色模式接近五五开,独立argmax会因微小概率变化在相邻位置切换,形成不自然花斑;独立抽样也可能破坏整件衣服一致性。局部边缘分布拟合得好,不代表最终联合图像结构正确。

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

      This causes problems. Let's return to our t-shirt example. We will use a per-pixel color classifier and see where it fails. Since the data distribution has two equally probable modes—teal and pink—the classifier will learn to place roughly equal probability mass on these two modes. As training data and time go to infinity, the classifier should recover the exact data distribution, but with finite data

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(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/c034.xhtml) - 结论依据:摘录中关于逐点决策可能破坏整体一致的定义、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:This causes problems. Let's return to our t-shirt example. We will use a per-pixel color classifier and see where it fails. Since the data distribution has two equally probable modes—teal and pink—the classifier will learn to place roughly equal probability mass on these two modes. As training data and time go to infinity, the classifier should recover the exact data distribution, but with finite data