← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8927.b · 生成模型:GPT-6

张量轴顺序必须写清

理解等级:中级

书中核按[输入通道,输出通道,高,宽]表示,其他实现可能采用不同顺序。只看相同维度数字无法判断其语义,特别在Cin=Cout时更易误用。调试应逐步标出输入、核、偏置和输出各轴,并统一从0或从1开始的索引约定。

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

      Notation for multichannel convolutions can get hard to keep track of, so let's spell out a few of the pieces here, which are also visualized in figure 24.5: xin[c1, :, :] is the c1-th channel of the input signal. The filter bank is Cout filters, [w[:, 0, :, :], …, w[:, Cout – 1, :, :]], each of which applies one convolutional filter per input

    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:摘录中关于张量轴顺序必须写清的定义、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:Notation for multichannel convolutions can get hard to keep track of, so let's spell out a few of the pieces here, which are also visualized in figure 24.5: xin[c1, :, :] is the c1-th channel of the input signal. The filter bank is Cout filters, [w[:, 0, :, :], …, w[:, Cout – 1, :, :]], each of which applies one convolutional filter per input