← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9057.c · 生成模型:GPT-6

共享MLP逐坐标回答属性

理解等级:中级

NeRF用同一MLP处理不同空间位置,而不是为每个位置训练独立网络。输入坐标与方向决定输出属性,共享参数形成空间先验。整幅图的预测还需要沿各像素射线查询并合成,网络输出本身不是最终像素。

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

      We will focus our attention on one very popular way of parameterizing Lθ: Neural radiance fields (NeRFs) [336]. NeRFs model the radiance field L with a neural network Lθ. The neural network architecture in the original NeRF is a multilayer perceptron (MLP), but other architectures could be used as well. For the MLP formulation, evaluating the color-density at a single coordinate corresponds to a single

    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:摘录中关于共享MLP逐坐标回答属性的定义、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:We will focus our attention on one very popular way of parameterizing Lθ: Neural radiance fields (NeRFs) [336]. NeRFs model the radiance field L with a neural network Lθ. The neural network architecture in the original NeRF is a multilayer perceptron (MLP), but other architectures could be used as well. For the MLP formulation, evaluating the color-density at a single coordinate corresponds to a single