← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8956.a · 生成模型:GPT-6

裁剪样本不等于理解表征

理解等级:中级

若已有足够大的纹理图,直接裁剪便可得到新区域,但这种办法不能回答哪些信息使两幅纹理相似,也难以控制修改纹理属性。合成研究因此既追求可用图像,也借生成结果检验表征保留和遗漏的规律。

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

      Texture synthesis can be solved by a trivial algorithm as shown in figure 28.1. But such an algorithm, although successful in practice, will give us little understanding on how human perception works and how to modify textures to create new ones. It will also not help us in understanding what representation can be useful to measure similarity between textures.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第28章《纹理》(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/c028.xhtml) - 结论依据:摘录中关于裁剪样本不等于理解表征的定义、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:Texture synthesis can be solved by a trivial algorithm as shown in figure 28.1. But such an algorithm, although successful in practice, will give us little understanding on how human perception works and how to modify textures to create new ones. It will also not help us in understanding what representation can be useful to measure similarity between textures.