← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9048.b · 生成模型:GPT-6

目标深度指导逆向重采样

理解等级:高级

对目标帧2像素p:P2=ẑ2(p)K2^−1p;将它以P1=R(1←2)P2+t(1←2)换到参考帧1,再求p1=π(K1P1),最后采样I1(p1)重建I2(p)。深度必须属于目标帧,位姿方向也必须是目标→参考;不能直接把前向运动不取逆套入。

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

      Given the depth at that location, recover the 3D world coordinates of the target point, P. To do this we will use the intrinsic camera matrix, assumed to be known, and the depth at that point: P = z(p)K–1p. As we do not have access to ground truth depth, we will use the function f1 to estimate depth: ẑ = f1(ℓ2). The coordinates of the

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第43章《单图深度估计》(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/c043.xhtml) - 结论依据:摘录中关于目标深度指导逆向重采样的定义、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:Given the depth at that location, recover the 3D world coordinates of the target point, P. To do this we will use the intrinsic camera matrix, assumed to be known, and the depth at that point: P = z(p)K–1p. As we do not have access to ground truth depth, we will use the function f1 to estimate depth: ẑ = f1(ℓ2). The coordinates of the