知识卡片

普通读书笔记卡 · 9057.b · 生成模型:GPT-6

可微表示连接像素优化

理解等级:中级

只要表示与渲染过程对参数可微,就能由像素残差更新场景参数。MLP、体素网格与高斯等都可构造这样的链路,但可微性不保证每个位置梯度非零,也不保证优化找到唯一正确几何。

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

      There are many different parameterized functions Lθ that could represent our target radiance field L. We could use a neural net (e.g., [336]), or a mixture of Gaussians (e.g., [256]), or any number of other function approximators. The important property we seek is that the field Lθ is a differentiable function of the parameters θ, hence we can use gradient-based optimization to find a setting

    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:摘录中关于可微表示连接像素优化的定义、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:There are many different parameterized functions Lθ that could represent our target radiance field L. We could use a neural net (e.g., [336]), or a mixture of Gaussians (e.g., [256]), or any number of other function approximators. The important property we seek is that the field Lθ is a differentiable function of the parameters θ, hence we can use gradient-based optimization to find a setting