← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9052.a · 生成模型:GPT-6

可见性权重只纳入真实观测

理解等级:中级

以v_ij标记点i是否在图j有有效观测,仅对v_ij=1的项计算重投影损失。未观测不等于像素坐标为零;缺失数据应通过掩码排除。这里通常是匹配得到的观测存在性,不是仅凭三维预测决定是否可见,也不能用掩码任意删除难例。

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

      To have the final error we still need to add one more factor. We add a visibility term, , to account for 3D points that are only be visible in a subset of the views. We will use the binary variable to indicate if point i is seeing from camera j. For all the points that are not visible in a view we set ,

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第44章《多视图几何与运动恢复结构》(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/c044.xhtml) - 结论依据:摘录中关于可见性权重只纳入真实观测的定义、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:To have the final error we still need to add one more factor. We add a visibility term, , to account for 3D points that are only be visible in a subset of the views. We will use the binary variable to indicate if point i is seeing from camera j. For all the points that are not visible in a view we set ,