← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9054.d · 生成模型:GPT-6

鲁棒核降低离群残差影响

理解等级:中级

束调整常将平方重投影误差换成Huber等鲁棒代价,降低少量大残差对结果的支配。它需要合理尺度与初值,也不保证任意错误匹配被识别或全部局部极小消失。几何验证、轨迹清理与稳健优化是互补步骤。

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

      To find the final set of 3D points, the camera poses and the 3D locations of the keypoints are then further refined by bundle adjustment (camera intrinsics are refined too, if needed). The reprojection error loss is usually rewritten using a robust cost function (typically the Huber kernel).

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第44章《多视图几何与运动恢复结构》(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/c044.xhtml) - 结论依据:摘录中关于鲁棒核降低离群残差影响的定义、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:To find the final set of 3D points, the camera poses and the 3D locations of the keypoints are then further refined by bundle adjustment (camera intrinsics are refined too, if needed). The reprojection error loss is usually rewritten using a robust cost function (typically the Huber kernel).