Convolutional Neural Networks
Computer Vision Problems:Image Classification;Object detection;Neural Style Transfer
Edge Detection
Vertical edge detection
垂直边缘filter \(\left[ \begin{matrix} 1 & 0 & -1 \\ 1 & 0 & -1 \\ 1 & 0 & -1 \end{matrix} \right] \tag{1}\) Horizontal edge detection
水平边缘filter \(\left[ \begin{matrix} 1 & 1 & 1 \\ 0 & 0 & 0 \\ -1 & -1 & -1 \end{matrix} \right] \tag{2}\)
Padding
- Valid Conv:对一幅\((n,n)\)的图像用\((f,f)\)的滤波器卷积,得到的结果图像尺寸是\((n-f+1,n-f+1)\)
- Same Conv(pad):对一幅\((n,n)\)的图像先pad到\((n+2,n+2)\),p=2,再与\((f,f)\)的滤波器卷积,得到的结果仍然是\((n,n)\)
Stride Convolutions
对一幅\((n*n)\)的图像,用\((f*f)\)的滤波器做padding = p,stride=s的卷积,得到的结果是 \((\frac{n+2p-f}{s}+1,\frac{n+2p-f}{s}+1)\)
卷积?互相关?
高维卷积
例如3维图像与3维滤波器卷积\((6,6,3)*(3,3,3) = (4,4)\)
\((n,n,n_c) * (f,f,n_c) = (n-f+1,n-f+1)...n_c\)应保持一致
convolutional network
- Convolution
- Pooling
- Fully connected
Pooling
- Max pooling:取区域内最大值
- Avg pooling:取区域内平均值
Hyperparameters:
- f:filter size
- s:stride
- Max or average pooling