← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9066.c · 生成模型:GPT-6

代价体可用流式计算

理解等级:中级

完整保存每像素每候选位移代价会产生约NM(2L+1)²规模的存储。可分块处理、逐位移更新最佳代价或复用计算,减少内存而保持同一搜索目标。优化前应保留一致的并列规则和边界处理。

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

      Algorithm 46.1: Patch matching motion estimation. The algorithm starts by chopping the two frames into overlapping patches. Then, for every patch from the first frame, we compute the distance to all the nearby patches in frame 2. Finally, for each input patch we select the closest patch from frame 2 and we record the relative displacement between the two patches. The pseudocode can be rearranged

    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:摘录中关于代价体可用流式计算的定义、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:Algorithm 46.1: Patch matching motion estimation. The algorithm starts by chopping the two frames into overlapping patches. Then, for every patch from the first frame, we compute the distance to all the nearby patches in frame 2. Finally, for each input patch we select the closest patch from frame 2 and we record the relative displacement between the two patches. The pseudocode can be rearranged