← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9080.b · 生成模型:GPT-6

聚合代价利用邻域证据

理解等级:中级

代价聚合网络将局部候选相似度转为更一致的位移估计,可结合邻域纹理、边界和语义。聚合不是单纯对每像素所有代价值求平均;输出需保留候选差异并选出或回归两个运动分量。

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

      Train and apply a CNN to aggregate (process) the costs over the cost volume in order to estimate a single best optical flow for each pixel position. Use a coarse-to-fine estimation procedure where optical flow estimated at a coarse scale is used to warp the features at a finer scale to compute a refined cost volume. Then, estimate an update to the optical flow to

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第49章《学习运动估计》(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/c049.xhtml) - 结论依据:摘录中关于聚合代价利用邻域证据的定义、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:Train and apply a CNN to aggregate (process) the costs over the cost volume in order to estimate a single best optical flow for each pixel position. Use a coarse-to-fine estimation procedure where optical flow estimated at a coarse scale is used to warp the features at a finer scale to compute a refined cost volume. Then, estimate an update to the optical flow to