知识卡片
干扰因素由任务定义
光照和视角对物体身份识别可能是干扰因素,对估计照明或相机姿态却是目标信息。鼓励表示对某种变换不变,相当于主动丢弃相关差异。压缩并不自动知道该丢什么,错误的不变性会伤害下游任务。
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
- · 表示学习
原书出处 · 1 处
- · 查看原书依据
2. Compression is a way to achieve invariance to nuisance factors. Depending on what the representation will be used for, nuisances might include camera noise, lighting conditions, and viewing angle. If we can factor out those nuisances and discard them from our representation, then the representation will be more useful for tasks like object recognition where camera and lighting may change without affecting the semantics
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第30章《表示学习》(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/c030b.xhtml)
- 结论依据:摘录中关于干扰因素由任务定义的定义、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:2. Compression is a way to achieve invariance to nuisance factors. Depending on what the representation will be used for, nuisances might include camera noise, lighting conditions, and viewing angle. If we can factor out those nuisances and discard them from our representation, then the representation will be more useful for tasks like object recognition where camera and lighting may change without affecting the semantics