CNN

Hello, I work on CNN , I use scipy library for the implementation .
My problem in the concatenation step
for example ( for i range (0,3)), I obtain 3 pooling maps , each pooling map is of size (2*2).But when I use flatten equation during the concatenation step ,In fact I must get a vector of size (12*1),but just I obtain a vector of size (4*1) corresponding to the last pooling .

Please who can help me:

Code:

import numpy as np,sys,array
from scipy.ndimage.filters import maximum_filter
import skimage.measure
from scipy.signal import convolve2d


x = np.array([[1, 2, 2,0],
    [1, 3, 1,1],
    [2, 0, 1,2],
    [0, 4, 2,1]])
x_with_pad = np.pad(x, (1, 1), mode='constant', constant_values=0)

class neuralNetwork(object):

     
    def __init__(self ):
     
        self.W1 = np.random.rand(2,4)-0.5
        self.W2 = np.random.rand(2,2)-0.5
        self.W3 = np.random.rand(1,2)-0.5
        self.lr = 0.3
       
    def forward(self, x_with_pad ):
        
        for i in range(0,3):
          self.F= np.random.rand(3,3)
          print("Filter is=",self.F )
          self.C = convolve2d(x_with_pad,self.F,mode='valid')   
          print("Convolution is=",self.C )
          self.Cp = self.ReLu(self.C)
          self.P= skimage.measure.block_reduce(self.Cp, (2,2), np.max)
          print("pooling is=",self.P )
        self.concatenation= self.P.flatten()
        print("concat is=",self.concatenation)
        self.H1 = np.dot(self.W1, self.concatenation)
        self.Y1 = self.sigmoid(self.H1)
        self.H2 = np.dot( self.W2,self.Y1)
        self.Y2 = self.sigmoid(self.H2)
        self.H3 = np.dot(self.W3,self.Y2)
        Y3 = self.sigmoid(self.H3)  
        return Y3

    def ReLu(self,x):
        mask = (x>0) * 1.0 
        return x * mask
    def d_ReLu(self,x):
        mask = (x>0) * 1.0 
        return  mask

    def sigmoid( self,s):
    # activation function
        return 1/(1+np.exp(-s))
  
    def sigmoidPrime(self,s):
    #derivative of sigmoid
        return s * (1 - s) 
    
NN = neuralNetwork()
for j in range(1): 
  NN.forward(x_with_pad )

CNN

CNN

Supprimer

Modérer

Signaler le message aux modérateurs