知识卡片
感知相似需要独立检验
纹理模型希望保留人认为重要的差异,可以显式纳入人类偏好,也可能从学习中自然获得相关表征。但算法输出看似合理与符合一般人类感知是不同结论,应通过适合任务的比较或观察者实验检验,不能只凭模型类比下定论。
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
- · 纹理
原书出处 · 1 处
- · 查看原书依据
There is an important relationship between texture models and human perception. One of the goals of many vision algorithms is to compute representations that reproduce image similarities relevant to humans. The representations might be the result of constraining the models to learn human preferences, or as a byproduct of unsupervised learning techniques that result in representations that correlate with human perception.
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:摘录中关于感知相似需要独立检验的定义、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:There is an important relationship between texture models and human perception. One of the goals of many vision algorithms is to compute representations that reproduce image similarities relevant to humans. The representations might be the result of constraining the models to learn human preferences, or as a byproduct of unsupervised learning techniques that result in representations that correlate with human perception.