← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8933.a · 生成模型:GPT-6

深层语义是经验趋势

理解等级:中级

分类网络常逐层减空间分辨率、增通道数,浅层响应边缘纹理,深层与类别语义关联更强。但通道不保证一一对应类别,也并非每层都必然降采样。分析具体模型应从实际形状和响应出发,避免把常见趋势当成架构定义。

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 a few things to notice in this figure. First, the spatial resolution of the feature maps get lower as we go deeper into the network, and the number of channels increases. This is common in CNNs: each layer downsamples and adds channels to partially compensate for the reduction in resolution. Second, while on the first layer the feature maps are sensitive to basic

    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:摘录中关于深层语义是经验趋势的定义、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 a few things to notice in this figure. First, the spatial resolution of the feature maps get lower as we go deeper into the network, and the number of channels increases. This is common in CNNs: each layer downsamples and adds channels to partially compensate for the reduction in resolution. Second, while on the first layer the feature maps are sensitive to basic