知识卡片
互斥分类在概率单纯形上
单标签互斥设定用K个非负概率且总和为一,形成K−1维单纯形,常用softmax输出。原文0>ŷ_c>1不可能,应为0≤ŷ_c≤1;多对象可共存场景不能仅因接口方便而强制成互斥标签。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 目标识别
原书出处 · 1 处
- · 查看原书依据
One common assumption is that only one class, among the K possible classes, is present in the image. This is typical in settings where most of the images contain a single large object. Multiclass classification with exclusive classes results is given in the following constraint:
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:摘录中关于互斥分类在概率单纯形上的定义、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:One common assumption is that only one class, among the K possible classes, is present in the image. This is typical in settings where most of the images contain a single large object. Multiclass classification with exclusive classes results is given in the following constraint: