← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9092.c · 生成模型:GPT-6

实例分割为每个对象建掩码

理解等级:中级

实例分割输出对象集合,每个实例有独立类别与像素掩码,因而能区分同类对象并支持计数。通常可见掩码不包含被遮挡的完整形状;处理重叠不等于已恢复每个物体的不可见背面。

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

      The difference from semantic segmentation is that if there are K objects of the same type, we will output K masks; conversely, semantic segmentation has no way to telling us how many objects of the same type there are, nor can it delineate between two objects of the same type that overlap; it just gives us an aggregate region of pixels that are all the

    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:摘录中关于实例分割为每个对象建掩码的定义、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:The difference from semantic segmentation is that if there are K objects of the same type, we will output K masks; conversely, semantic segmentation has no way to telling us how many objects of the same type there are, nor can it delineate between two objects of the same type that overlap; it just gives us an aggregate region of pixels that are all the