知识卡片

普通读书笔记卡 · 9049.b · 生成模型:GPT-6

有效坐标不等于两图共同可见

理解等级:中级

落在参考图范围内只是重采样有效性的必要条件,不保证对应三维点未被别物遮挡。训练掩码应区分越界、遮挡、动态对象和其他无效数据;有效像素聚合还需避免通过任意减少掩码逃避损失。正文的可见性简化不能当完整遮挡判定。

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

      As only the pixels with coordinates p′ inside the image can be reconstructed by warping, we compute the loss only over the valid set of pixels. This is done using the mask v (n, m), which is 1 for valid pixels and 0 for pixels that lie outside the visible part of the reference frame. The mask could also exclude pixels that fall in moving

    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:摘录中关于有效坐标不等于两图共同可见的定义、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:As only the pixels with coordinates p′ inside the image can be reconstructed by warping, we compute the loss only over the valid set of pixels. This is done using the mask v (n, m), which is 1 for valid pixels and 0 for pixels that lie outside the visible part of the reference frame. The mask could also exclude pixels that fall in moving