知识卡片
识别表示按任务逐步丰富
分类、边界框检测、语义分割与实例分割组合了类别、空间范围和身份信息,各有表达限制。它们不是同一性能指标上的简单高低排序,最合适的输出仍取决于下游是否需要计数、像素归属或实例行动。
结构图:
flowchart TD
A[识别所需输出] --> B[整图类别存在性]
A --> C[实例类别与边界框]
A --> D[逐像素语义类别]
C --> E[实例类别与独立掩码]
D --> E
E --> F[仍需额外几何与物理信息]
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
Figure 50.24 shows a pictorial summary of the different methods we have seen in this chapter to represent objects in images. The four different approaches we have described use a common set of building blocks: image classifiers, region proposals, pixelwise classifiers, and bounding box regression.
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:摘录中关于识别表示按任务逐步丰富的定义、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 50.24 shows a pictorial summary of the different methods we have seen in this chapter to represent objects in images. The four different approaches we have described use a common set of building blocks: image classifiers, region proposals, pixelwise classifiers, and bounding box regression.