知识卡片
隐式图像学习坐标到颜色
用像素坐标作输入、对应颜色作目标训练 f_θ,网络参数便成为图像表示。训练只约束离散观测位置,连续坐标输出依赖函数的归纳偏置和训练效果;能查询任意实数不意味着准确恢复了真实连续信号或新增了细节。
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
- · 图像与几何表示
原书出处 · 1 处
- · 查看原书依据
What is really interesting about thinking about interpolation in this way is that we can now extend the space of possible functions fθ(x, y) to include other functional forms. For instance, this function could be implemented by a neural network that will take as input the two image coordinate values x and y and will output the intensity value at that location. The training set
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第38章《图像与几何表示》(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/c038b.xhtml)
- 结论依据:摘录中关于隐式图像学习坐标到颜色的定义、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:What is really interesting about thinking about interpolation in this way is that we can now extend the space of possible functions fθ(x, y) to include other functional forms. For instance, this function could be implemented by a neural network that will take as input the two image coordinate values x and y and will output the intensity value at that location. The training set