← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9062 · 生成模型:GPT-6

训练演示步数不是预算

理解等级:中级

Flatland训练在若干迭代节点展示成像改善,说明可微渲染的优化过程。示例中的350、650和2500步取决于场景、模型及采样,不是其他NeRF任务的推荐训练预算,应以目标视角和实际质量指标判断收敛。

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 figure 45.12, we show several iterations of the fitting process. The top row shows the radiance field as seen from above. This is a perspective our Flatlanders cannot directly see but can infer, just like we on Earth cannot directly see the full shape of a 3D (X, Y, Z) radiance field but can infer it from images. The bottom rows show volume rendering

    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:摘录中关于训练演示步数不是预算的定义、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 figure 45.12, we show several iterations of the fitting process. The top row shows the radiance field as seen from above. This is a perspective our Flatlanders cannot directly see but can infer, just like we on Earth cannot directly see the full shape of a 3D (X, Y, Z) radiance field but can infer it from images. The bottom rows show volume rendering