← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9003.c · 生成模型:GPT-6

搜索结果继承采集偏差

理解等级:中级

书中 mug 搜索样本多为白底、完整可见、典型侧视的杯子,缺少俯视、手部遮挡、柜内和水槽场景。搜索能扩大数量,却仍受查询词、来源和排序机制筛选。检查采集协议而不只统计类别数,才能发现真实部署需要的变化缺口;图中结果是历史示例。

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

      Figure 35.10 shows the first 22 images returned by Google for the query “mug.” It looks far from a diverse and unbiased collection of images. As you can appreciate these pictures of mugs look very different from the visual settings in which one usually encounters mugs (inside cabinets, on top of a table, inside the dishwasher, or dirty inside the sink). Instead the images that

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第35章《数据偏差与分布漂移》(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/c035b.xhtml) - 结论依据:摘录中关于搜索结果继承采集偏差的定义、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:Figure 35.10 shows the first 22 images returned by Google for the query “mug.” It looks far from a diverse and unbiased collection of images. As you can appreciate these pictures of mugs look very different from the visual settings in which one usually encounters mugs (inside cabinets, on top of a table, inside the dishwasher, or dirty inside the sink). Instead the images that