知识卡片

普通读书笔记卡 · 9089.a · 生成模型:GPT-6

背景框只监督分类职责

理解等级:中级

检测总损失将分类与定位结合,但只对有目标的正例计算框回归:L_cls+λ1(y_i≠0)L_loc。公式50.17的非零条件正确,正文写1(y_i=0)则反了;背景没有可定义的对象框,不能用它计算任意回归目标。

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

      where the indicator function 𝟙(yi = 0) sets to zero the location loss for the bounding boxes that do not overlap with any of the ground truth bounding boxes. The parameter λ can be used to balance the relative strength of both losses. This loss makes it possible to train the whole detection algorithm end-to-end. It is possible to train the localization and the classification

    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:摘录中关于背景框只监督分类职责的定义、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:where the indicator function 𝟙(yi = 0) sets to zero the location loss for the bounding boxes that do not overlap with any of the ground truth bounding boxes. The parameter λ can be used to balance the relative strength of both losses. This loss makes it possible to train the whole detection algorithm end-to-end. It is possible to train the localization and the classification