← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9017.a · 生成模型:GPT-6

变换点集后仍需重采样

理解等级:中级

旋转后的像素集合通常落在非整数、不规则位置。若还要使用规则网格上的强度卷积,必须重新栅格化并处理插值与滤波;仅更新坐标不会自动生成可显示图像,也不会保留所有原网格邻域的关系。

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

      However, note that to perform a convolution on the output intensity will require translating the position encoding back into an image on a rectangular grid. For instance, after a rotation, the locations for the intensity values will change and the pixels will not lie on a rectangular grid anymore. The convolution kernels for the intensity channel will have to be transformed too.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第38章《图像与几何表示》(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/c038b.xhtml) - 结论依据:摘录中关于变换点集后仍需重采样的定义、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:However, note that to perform a convolution on the output intensity will require translating the position encoding back into an image on a rectangular grid. For instance, after a rotation, the locations for the intensity values will change and the pixels will not lie on a rectangular grid anymore. The convolution kernels for the intensity channel will have to be transformed too.