← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8912.c · 生成模型:GPT-6

上采样不等于超分辨率

理解等级:初级

本章上采样通过插值增加样本数,目标是用更密的网格表达已有信息。超分辨率还要恢复或预测低分辨率中缺失的细节,属于更强的逆问题。输出像素变多不会自动使观测信息增多,图像看起来平滑也不证明恢复了真实高频。

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

      Upsampling requires increasing the resolution of an image from a low resolution image. This is a very challenging task that requires recovering information that is not available on the low resolution image. The problem of making up new details will be called super-resolution. In this section we will focus on the simpler problem of creating an image that has k times more pixels along each

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第21章《图像下采样与上采样》(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/c021.xhtml) - 结论依据:摘录中关于上采样不等于超分辨率的定义、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:Upsampling requires increasing the resolution of an image from a low resolution image. This is a very challenging task that requires recovering information that is not available on the low resolution image. The problem of making up new details will be called super-resolution. In this section we will focus on the simpler problem of creating an image that has k times more pixels along each