知识卡片

普通读书笔记卡 · 9092 · 生成模型:GPT-6

分割均值IoU关注各类覆盖

理解等级:中级

像素准确率容易由路面、天空等大类主导,先对每类求TP/(TP+FP+FN)再取平均能提高小类的权重。需明确缺失类和忽略像素规则;它不是完整的尺度无关保证,小物体同样像素边界误差仍会损失较大IoU。

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 order to evaluate the quality of the segmentation, we can measure the percentage of correctly labeled pixels. However, this measure will be dominated by the large objects (sky, road, etc.). A more informative measure is the average IoU across classes. The IoU is a measure that is invariant to scale and will provide a better characterization of the overall segmentation quality across all classes.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第50章《目标识别》(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/c050b.xhtml) - 结论依据:摘录中关于分割均值IoU关注各类覆盖的定义、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 order to evaluate the quality of the segmentation, we can measure the percentage of correctly labeled pixels. However, this measure will be dominated by the large objects (sky, road, etc.). A more informative measure is the average IoU across classes. The IoU is a measure that is invariant to scale and will provide a better characterization of the overall segmentation quality across all classes.