知识卡片

普通读书笔记卡 · 8996.a · 生成模型:GPT-6

联合量化类别数指数增长

理解等级:中级

若每像素两个颜色通道各分k档,N像素联合配置共有k的2N次方种。把每种完整图像作为一个类别很快不可行。逐像素分类降低输出数量,却施加条件独立或需要额外结构建模;向量量化改变码本设计,也不免费消除全部高维困难。

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

      The real problem comes about when predicting more than one pixel. To quantize the ab-values of N pixels requires k2N classes, again assuming we want a resolution of k for the a and b value of each pixel. For a 256 × 256 resolution image, the number of classes required is astronomical (for k = 10, the number is a one followed by over 100,000

    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:摘录中关于联合量化类别数指数增长的定义、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:The real problem comes about when predicting more than one pixel. To quantize the ab-values of N pixels requires k2N classes, again assuming we want a resolution of k for the a and b value of each pixel. For a 256 × 256 resolution image, the number of classes required is astronomical (for k = 10, the number is a one followed by over 100,000