知识卡片

普通读书笔记卡 · 9088.a · 生成模型:GPT-6

级联让早期快速拒绝背景

理解等级:中级

级联以多个判别阶段逐渐筛选候选,让多数背景在廉价早期退出。每层错拒会累积降低召回,因此早期应兼顾高通过率与足够排除率;层数越多、单层越准确,并不自动带来最优总成本。

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

      Trying to localize an object in an image is like finding a needle in a haystack: the object is usually small and might be surrounded by a complex background. Selective search reduced the complexity of the search by dividing the search in two steps: a first, fast, and cheap classification function that detects good candidate locations; and a second, slow, and expensive classification function capable

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第50章《目标识别》(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/c050b.xhtml) - 结论依据:摘录中关于级联让早期快速拒绝背景的定义、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:Trying to localize an object in an image is like finding a needle in a haystack: the object is usually small and might be surrounded by a complex background. Selective search reduced the complexity of the search by dividing the search in two steps: a first, fast, and cheap classification function that detects good candidate locations; and a second, slow, and expensive classification function capable