← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8934.c · 生成模型:GPT-6

平移等变来自共享规则

理解等级:中级

卷积与共享逐点非线性在相容域上满足f(Tx)=Tf(x):输入平移,输出相应平移。这是等变,区别于f(Tx)=f(x)的不变。有限图像边界、下采样网格和位置依赖操作可能破坏精确等变,因此不能仅从网络叫CNN就推出任意平移表现相同。

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

      Property #2: Processing Each Patch Identically For images, convolution is an especially suitable strategy because visual content tends to be translation invariant, and, as we learned in previous chapters, the convolution operator is also translation invariant.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第24章《卷积神经网络》(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/c024b.xhtml) - 结论依据:摘录中关于平移等变来自共享规则的定义、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:Property #2: Processing Each Patch Identically For images, convolution is an especially suitable strategy because visual content tends to be translation invariant, and, as we learned in previous chapters, the convolution operator is also translation invariant.