知识卡片

结构图卡 · 8921.d · 生成模型:GPT-6

运动能量网络分层计算

理解等级:中级

结构图:

flowchart LR
 A[视频体] --> B[时空正弦余弦滤波器组]
 B --> C[逐路平方]
 C --> D[正交对求和]
 D --> E[按候选速度平面聚合]
 E --> F[可选响应归一化]
 F --> G[比较速度候选]

平方与归一化让完整架构非线性。图22.17实际输出平方和,若另行开根,应在聚合前后明确位置;零总响应时不能直接除以总和。

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 an illustration, figure 22.17 shows one possible architecture to create velocity-selective units. The first layer is composed by space-time Gabor filters (cosine and sine), which are frequency-selective units. Here we represent the impulse response of each filter by a small x – y – t cube. For each quadrature pair we compute the amplitude. Then amplitude outputs are combined according to form different planes

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第22章《滤波器组》(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/c022.xhtml) - 结论依据:摘录中关于运动能量网络分层计算的定义、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:As an illustration, figure 22.17 shows one possible architecture to create velocity-selective units. The first layer is composed by space-time Gabor filters (cosine and sine), which are frequency-selective units. Here we represent the impulse response of each filter by a small x – y – t cube. For each quadrature pair we compute the amplitude. Then amplitude outputs are combined according to form different planes