← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9078.a · 生成模型:GPT-6

重建损失必须统一帧方向

理解等级:中级

前向流可用Σ|I₂(p+f(p))−I₁(p)|²比较有效像素。式48.12把帧下标和时间写法混用,使用时应先定义参考帧与采样帧,明确插值及越界掩码,避免把反向warp误认为前向splat。

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

      The gradient-based approach is efficient but it only provides an approximation to the motion field because it ignores higher order terms in the Taylor expansion in equation (48.4). A different approach consists of directly minimizing the photometric reconstruction error:

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第48章《光流估计》(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/c048.xhtml) - 结论依据:摘录中关于重建损失必须统一帧方向的定义、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:The gradient-based approach is efficient but it only provides an approximation to the motion field because it ignores higher order terms in the Taylor expansion in equation (48.4). A different approach consists of directly minimizing the photometric reconstruction error: