知识卡片

普通读书笔记卡 · 9086.c · 生成模型:GPT-6

分类捷径会伪装对象理解

理解等级:中级

整图标签允许模型靠背景、局部部件或采集相关性作出正确分类,而无需完整定位对象。高分类分数因此不证明掌握了对象结构。需用背景变化、局部遮挡或定位等独立任务检验所依赖的信息。

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

      Is it possible to classify an image without localizing the object? How can we answer the question “Is object class c present in the image?” without localizing the object? One possibility is that the function f has learned to localize the object internally, but the information about the location is not being recorded in the output. Another possibility is that the function has learned to

    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:摘录中关于分类捷径会伪装对象理解的定义、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:Is it possible to classify an image without localizing the object? How can we answer the question “Is object class c present in the image?” without localizing the object? One possibility is that the function f has learned to localize the object internally, but the information about the location is not being recorded in the output. Another possibility is that the function has learned to