知识卡片
图像分类不定位也不计数
整图分类回答某类别是否出现,通常在固定有限词表上给出分数。它没有输出该对象在哪里或有几个,即使网络内部利用了局部信息,也不能从分类接口直接读取可靠位置与实例数。
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
Image classification typically assumes that there is a closed vocabulary of object classes with a finite number of predefined classes. If our vocabulary contains K classes, we can loop over the classes and ask the same question for each class. This task does not try to localize the object in the image, or to count how many instances of the object are present.
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:摘录中关于图像分类不定位也不计数的定义、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:Image classification typically assumes that there is a closed vocabulary of object classes with a finite number of predefined classes. If our vocabulary contains K classes, we can loop over the classes and ask the same question for each class. This task does not try to localize the object in the image, or to count how many instances of the object are present.