知识卡片

普通读书笔记卡 · 9066.b · 生成模型:GPT-6

边界处理需覆盖完整块

理解等级:中级

边界像素的候选中心与候选块都可能越出图像,填充或有效掩码需要同时考虑搜索半径L和块半径s。仅将中心限制到离边界L处不一定足够;填充值也会影响匹配,应明示不可靠区域。

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

      Note that padding must be applied if we want to compute the output optical flow near the image boundaries. Algorithm 46.1 could be written more compactly but we prefer this form for its clarity. The following images shows the matching result for one input location. In this example, s = 5 (patch size of 11 × 11 pixels), and L = 16 (search window of

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第46章《运动估计》(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/c046b.xhtml) - 结论依据:摘录中关于边界处理需覆盖完整块的定义、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:Note that padding must be applied if we want to compute the output optical flow near the image boundaries. Algorithm 46.1 could be written more compactly but we prefer this form for its clarity. The following images shows the matching result for one input location. In this example, s = 5 (patch size of 11 × 11 pixels), and L = 16 (search window of