← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9046.a · 生成模型:GPT-6

监督深度是带来源的回归

理解等级:中级

监督训练用图像与对应深度图最小化预测误差。目标可以来自RGB-D、LiDAR、立体或多视图重建,不同来源有不同噪声、缺失和尺度约束。训练前检查对齐、单位、有效区域和相机模式,不把所有标注统称为无误差真值。

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

      In the supervised setting we will assume that we have a training set that contains N images, ℓ(i)[n, m], and their associated depth maps, z(i)[n, m]. In this formulation, depth estimation becomes a pixelwise regression problem. Usually z(i) and ℓ(i) are images of the same size. Our goal is to estimate depth at each pixel, ẑ(i)[n, m], from a single image using a learnable function,

    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:摘录中关于监督深度是带来源的回归的定义、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:In the supervised setting we will assume that we have a training set that contains N images, ℓ(i)[n, m], and their associated depth maps, z(i)[n, m]. In this formulation, depth estimation becomes a pixelwise regression problem. Usually z(i) and ℓ(i) are images of the same size. Our goal is to estimate depth at each pixel, ẑ(i)[n, m], from a single image using a learnable function,