← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9047 · 生成模型:GPT-6

全局对齐让损失耦合像素

理解等级:中级

损失中的α来自全图平均误差,所以即使最终写成逐像素求和,各项也通过α相互联系,不能称为完全独立比较。归一化方式、有效像素数与对齐参数会影响优化和评估;训练与测试是否允许用真值拟合尺度必须明确说明。

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

      where . This loss compares the estimated and ground truth depth at each pixel independently. In order to get a more globally consistent reconstruction one can also introduce in the loss the depth derivatives, although ground truth derivatives might be noisy depending on how the depth has been collected. Optimizing depth reconstruction might over emphasize the errors for points that large depth values. One way

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第43章《单图深度估计》(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/c043.xhtml) - 结论依据:摘录中关于全局对齐让损失耦合像素的定义、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:where . This loss compares the estimated and ground truth depth at each pixel independently. In order to get a more globally consistent reconstruction one can also introduce in the loss the depth derivatives, although ground truth derivatives might be noisy depending on how the depth has been collected. Optimizing depth reconstruction might over emphasize the errors for points that large depth values. One way