知识卡片

普通读书笔记卡 · 8928.c · 生成模型:GPT-6

级联空洞核增加表达范围

理解等级:中级

级联不同扩张率的短核,可以用较少参数形成跨多个尺度的等效大核;若中间没有非线性,等效核就是各扩张核的卷积。三层3×3、扩张率1/2/4的跨度为1+2(1+2+4)=15,参数27个,仍不是任意15×15核都能表示。

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

      There are ways in which dilation can be used to increase the family of useful filters. For instance, by composing three convolutions with dk = 1, dk = 2, and dk = 4 together (figure 24.12), one can create a kernel that can switch during learning between high and low spatial frequencies and small and large kernels.

    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:摘录中关于级联空洞核增加表达范围的定义、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:There are ways in which dilation can be used to increase the family of useful filters. For instance, by composing three convolutions with dk = 1, dk = 2, and dk = 4 together (figure 24.12), one can create a kernel that can switch during learning between high and low spatial frequencies and small and large kernels.