← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8985.c · 生成模型:GPT-6

像素顺序改变模型归纳偏好

理解等级:中级

将二维图像按行展开是一种有效自回归顺序,也可以选择其他固定顺序。链式分解本身总成立,但有限模型容量、可见上下文和位置编码随顺序而变化。训练与生成必须采用一致的像素及通道顺序。

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

      You might be wondering, how do we turn an image into a sequence of pixels? Good question! There are innumerable ways, but the simplest is often good enough: just vectorize the two-dimensional (2D) grid of pixels by first listing the first row of pixels, then the second row, and so forth. In general, any fixed ordering of the pixels into a sequence is actually valid,

    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:摘录中关于像素顺序改变模型归纳偏好的定义、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:You might be wondering, how do we turn an image into a sequence of pixels? Good question! There are innumerable ways, but the simplest is often good enough: just vectorize the two-dimensional (2D) grid of pixels by first listing the first row of pixels, then the second row, and so forth. In general, any fixed ordering of the pixels into a sequence is actually valid,