← 返回起始笔记位置

知识卡片

结构图卡 · 9061.a · 生成模型:GPT-6

训练链路连接场与像素

理解等级:中级

像素坐标与相机生成射线,射线采样得到空间点,场网络返回颜色密度,再由体渲染合成像素并计算残差。方向、采样距离与相机参数是链路必要输入;图上的简化串联不表示它们可以省略。

结构图:

flowchart LR
 A[像素与相机] --> B[世界射线]
 B --> C[采样位置与间距]
 C --> D[颜色与密度场]
 D --> E[体渲染合成]
 E --> F[观测像素残差]
 F --> G[更新场参数]
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

      Given Lθ, we will render an image simply using volume rendering as described previously. The entire computation graph is shown in figure 45.11. Notice that all the modules are differentiable, which will be critical for our next step, fitting the parameters to data.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第45章《辐射场》(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/c045.xhtml) - 结论依据:摘录中关于训练链路连接场与像素的定义、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:Given Lθ, we will render an image simply using volume rendering as described previously. The entire computation graph is shown in figure 45.11. Notice that all the modules are differentiable, which will be critical for our next step, fitting the parameters to data.