← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8927.a · 生成模型:GPT-6

卷积参数数由通道决定

理解等级:中级

普通二维密集卷积的空间核大小K×K、输入通道Cin、输出通道Cout时,权重数为Cin·Cout·K²,另有Cout个输出偏置。RGB经3×3核得到96路,每个输出核27个权重,共2592个,是否保留128×128空间大小还取决于padding。

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

      Multi-Input, Multi-Output Putting both of the above together, we can define a general convolutional layer that maps a signal with Cin input channels to a signal with Cout output channels. Here is what this looks like for an image , where c2 indexes the output channel, with c2 ∈ {0, …, Cout – 1}:

    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:摘录中关于卷积参数数由通道决定的定义、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:Multi-Input, Multi-Output Putting both of the above together, we can define a general convolutional layer that maps a signal with Cin input channels to a signal with Cout output channels. Here is what this looks like for an image , where c2 indexes the output channel, with c2 ∈ {0, …, Cout – 1}: