知识卡片
逐点网络不等于空间卷积
把共享MLP应用在规则网格的各点,可类比逐点的1×1卷积,关键是各位置共享映射。这个类比不意味着网络自动混合相邻网格内容;空间关系来自输入坐标、编码及共享参数,而不是邻域卷积核。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 辐射场
原书出处 · 1 处
- · 查看原书依据
Often, however, we want to evaluate the color-density at multiple coordinate locations. To create an explicit representation of the field's values at a grid of coordinates, we can query the field on this grid. In this usage, the NeRF Lθ looks like a convolutional neural network (CNN) with 1x1 filters, as this is the architecture that results from applying an MLP to each position in
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:摘录中关于逐点网络不等于空间卷积的定义、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:Often, however, we want to evaluate the color-density at multiple coordinate locations. To create an explicit representation of the field's values at a grid of coordinates, we can query the field on this grid. In this usage, the NeRF Lθ looks like a convolutional neural network (CNN) with 1x1 filters, as this is the architecture that results from applying an MLP to each position in