知识卡片
语义掩码不携带实例身份
语义分割把同类像素合并,因此相接同类对象缺少分离信息。即使两个连通区域分开,也未必等于两个对象,一个遮挡实例可分成多个区域;不能只靠连通分量得到可靠实例计数。
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
In this representation we have lost something that we had with the bounding boxes: this representation cannot count instances. The semantic segmentation representation, as formulated in this section, cannot separate two instances of the same class that are in contact; these will appear as a single segment with the same label.
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:摘录中关于语义掩码不携带实例身份的定义、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 this representation we have lost something that we had with the bounding boxes: this representation cannot count instances. The semantic segmentation representation, as formulated in this section, cannot separate two instances of the same class that are in contact; these will appear as a single segment with the same label.