← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8934.a · 生成模型:GPT-6

全卷积栈是非线性滑窗

理解等级:中级

由共享卷积和逐点非线性组成的全卷积网络,可把每个输出看成同一函数F作用于其输入感受野。重叠窗口共享中间计算,无须真的逐块重复运行。边界padding、步长、输入相关的全局归一化或其他全局交互,会改变该局部同构解释的适用范围。

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 core of CNNs are the convolutional layers, and in this section we will consider a CNN with only conv layers interleaved with pointwise nonlinearities. Such a CNN is sometimes called a fully convolutional network or FCN [305]. What we will show below is that a whole FCN is just another sliding image filter.

    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:摘录中关于全卷积栈是非线性滑窗的定义、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 core of CNNs are the convolutional layers, and in this section we will consider a CNN with only conv layers interleaved with pointwise nonlinearities. Such a CNN is sometimes called a fully convolutional network or FCN [305]. What we will show below is that a whole FCN is just another sliding image filter.