实现Pooling层的前向传播与反向传播

妖狐艹你老母 2023-07-10 05:24 22阅读 0赞

第一次学习，看的这个视频教程实在觉得不好，这个过程也不是很理解，以后我会结合相关书籍和视频课程继续学习，遇到好的一定会为大家推荐  
（这是我的一个不太理解的过程=^=！）  
以三层卷积为例：  
 d a t a → W 1 , b 1 c o n v 1 → r e l u → p o o l → W 2 , b 2 f c → W 3 , b 3 o u t data\\overset\{W1,b1\}\{\\rightarrow\}conv1\\rightarrow relu\\rightarrow pool\\overset\{W2,b2\}\{\\rightarrow\}fc\\overset\{W3,b3\}\{\\rightarrow\}out data→W1,b1conv1→relu→pool→W2,b2fc→W3,b3out

max-pooling前向传播：

def max_pool_forward_naive(x,pool_param):
        HH,WW=pool_param['pool_height'],pool_param['pool_width']
        s=pool_param['stride']
        N,C,H,W=x.shape
        H_new=1+(H-HH)/s
        W_new=1+(W-WW)/s
        out=np.zeros((N,C,H_new,W_new))
        for i in range(N):
            for j in range(C):
                for k in range(H_new):
                    for l in range(W_new):
                        window=x[i,j,k*s:HH+k*s,l*s:WW+l*s]
                        out[i,j,k,l]=np.max(window)
        
        cache=(x,pool_param)
        return out,cache

max-pooling反向传播：

def max_pool_backward_naive(dout,cache):
        x,pool_param=cache
        HH,WW=pool_param['pool_height'],pool_param['pool_weight']
        s=pool_param['stride']
        N,C,H,W=x.shape
        H_new=1+(H-HH)/s
        W_new=1+(W-WW)/s
        dx=np.zeros_like(x)
        for i in range(N):
            for j in range(C):
                for k in range(H_new):
                    for l in range(W_new):
                    	window=x[i,j,k*s:HH+k*s,l*s:WW+l*s]
                        m=np.max(window)
                        dx[i,j,k*s:HH+k*s,l*s:WW+l*s]=(window==m)*dout(i,j,k,l)
    
        return dx