← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8927.c · 生成模型:GPT-6

步长降低输出采样密度

理解等级:中级

步长决定卷积核应用位置之间的间隔,以降低激活图尺寸和计算量。精确输出尺寸还依赖输入长、有效核宽、padding和取整规则;N/s只是特定对齐条件下的简化。是否尺寸可整除与是否抗混叠是两项独立条件。

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

      Convolutional layers, as defined previously, maintain the spatial resolution of the signal they process. However, commonly it is sufficient, or even desirable, to output a lower resolution. This can be achieved with strided convolution:

    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:摘录中关于步长降低输出采样密度的定义、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:Convolutional layers, as defined previously, maintain the spatial resolution of the signal they process. However, commonly it is sufficient, or even desirable, to output a lower resolution. This can be achieved with strided convolution: