← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9058.d · 生成模型:GPT-6

参数化选择改变计算方式

理解等级:中级

规则体素、MLP和空间高斯可表示不同形式的场,它们在存储、查询与优化上的权衡不同。椭球形高斯示意不等于完整描述了某个3D Gaussian Splatting实现,也不要求将所有高斯权重解释为归一化概率混合。

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

      In principle Lθ could be other kinds of functions, including those that are not neural nets. The important property it needs to have is that it be differentiable, so that we can optimize its parameters to fit a given scene. One alternative is to parameterize the radiance fields using a voxel-based representation, as was explored in, e.g., [147]. Another alternative is to use a mixture

    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:摘录中关于参数化选择改变计算方式的定义、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:In principle Lθ could be other kinds of functions, including those that are not neural nets. The important property it needs to have is that it be differentiable, so that we can optimize its parameters to fit a given scene. One alternative is to parameterize the radiance fields using a voxel-based representation, as was explored in, e.g., [147]. Another alternative is to use a mixture