pytorch 방법 테스트 - 볼륨 (3차원)

import torch
import torch.nn as nn
import torch.autograd as autograd

m = nn.Conv3d(2, 1, 3, stride=2)
input = autograd.Variable(torch.randn(1, 2, 3, 5, 6))
output = m(input)

print("    ：")
print(input)
print("     ：")
print(m.weight)
print("     ：")
print(m.bias)

print("        ：")
print(output)
print("     ：")
print(output.size())

convBlockOne = 0
convBlockOneStride = 0
for i in range(3):
    for j in range(3):
        for k in range(3):
            #              
            convBlockOne += m.weight[0][0][i][j][k] * input[0][0][i][j][k] \
                            + m.weight[0][1][i][j][k] * input[0][1][i][j][k]\
            #              (  stride 2)
            convBlockOneStride += m.weight[0][0][i][j][k] * input[0][0][i][j+2][k] \
                            + m.weight[0][1][i][j][k] * input[0][1][i][j+2][k]
convBlockOne += m.bias[0]
convBlockOneStride += m.bias[0]
print("         ：")
print(convBlockOne)
print("             (  stride 2)   ：")
print(convBlockOneStride)

   ：

    ：
tensor([[[[[-0.2675,  0.0556, -1.9855,  1.7392, -0.6632,  0.1453],
           [ 0.7349,  0.9476,  0.5355, -0.5594,  2.1242,  1.4058],
           [ 1.6077,  0.4459, -0.9089, -1.0717,  0.9295, -0.4676],
           [ 0.5669,  2.0196, -0.5530,  0.5664, -0.0614, -0.5880],
           [-0.2644,  2.4068,  0.8025,  0.7653, -0.4151, -1.7002]],


          [[-0.7315,  0.5338, -0.8345, -0.5890, -0.3181,  0.1182],
           [-0.4932,  0.4393, -2.2104,  1.6638, -1.9973, -0.5577],
           [ 1.5920,  0.9191,  0.0097,  1.3372, -0.7334, -0.1244],
           [-0.3916, -0.7088,  0.2680, -0.1301, -1.3089, -1.6015],
           [-0.5146,  0.3306, -0.0946,  0.7855,  0.6388,  0.1586]],


          [[ 1.9393, -1.4019,  0.2299, -0.9965,  0.7452,  0.3587],
           [-1.3883, -1.1940,  0.8678, -0.5619,  0.0258, -0.4928],
           [-2.0170, -0.3001, -1.0647, -0.7010, -1.4591, -0.0430],
           [-0.2462,  0.1500,  0.1057,  0.4905,  0.0698,  0.4276],
           [ 1.3770, -0.3881,  0.5839, -1.2195,  0.1621,  1.7151]]],




         [[[-0.5271,  1.1861, -0.0419, -1.3207, -0.5641,  0.3916],
           [ 1.3733,  0.6005,  0.1389, -1.4528, -0.4616, -0.5287],
           [ 1.2038,  0.6351, -1.0303, -0.2564,  0.5147, -1.1273],
           [ 1.6099, -1.8206, -0.6026,  1.3047, -0.5663,  0.2222],
           [-0.3679, -0.2235,  0.4051,  1.3802,  0.7523, -0.5938]],


          [[-0.1804, -1.8722,  2.2938,  2.7384,  0.1001,  0.8501],
           [-0.1188,  1.1934, -1.4962, -2.2822,  0.7555,  0.3891],
           [ 2.1036, -0.8809,  0.1145, -0.8379,  0.2282,  2.1451],
           [ 1.4755, -2.1665,  0.7929, -0.0130, -0.3857, -0.0294],
           [-1.0380,  0.1234,  2.1946, -1.3495, -0.4712,  0.9102]],


          [[-1.0044, -1.5893, -1.2136, -0.0563,  2.0776,  0.9950],
           [-0.9101,  0.4796,  0.3117,  0.1360, -2.4304,  0.4099],
           [ 0.0862,  1.0328,  0.9845, -0.9481,  1.9113,  1.3451],
           [-0.1578, -0.4626,  0.3089, -0.9687, -1.5955,  0.8957],
           [ 0.6446, -0.0345,  0.2948,  0.8574, -0.5022, -0.5117]]]]])
     ：
Parameter containing:
tensor([[[[[ 0.0134,  0.1058, -0.0204],
           [-0.1219,  0.0881, -0.0455],
           [-0.0813,  0.0803,  0.0774]],


          [[-0.0703, -0.0120, -0.0227],
           [-0.0820,  0.0459, -0.0942],
           [ 0.0043, -0.0486, -0.1331]],


          [[-0.1315, -0.0889,  0.1048],
           [-0.1034,  0.0881,  0.0259],
           [-0.0182,  0.0609,  0.0272]]],




         [[[-0.1038,  0.0391,  0.0325],
           [-0.0242,  0.1138,  0.0522],
           [ 0.0649,  0.0600, -0.0476]],


          [[ 0.1099, -0.0557,  0.0714],
           [-0.0997, -0.0301, -0.0354],
           [-0.1291, -0.0227,  0.0915]],


          [[-0.0529, -0.0907,  0.0566],
           [-0.0386, -0.0340, -0.0872],
           [-0.0612, -0.0455, -0.0208]]]]])
     ：
Parameter containing:
tensor(1.00000e-02 *
       [-7.2022])
        ：
tensor([[[[[ 0.4036,  0.7170],
           [ 0.4118,  0.1331]]]]])
     ：
torch.Size([1, 1, 1, 2, 2])
         ：
tensor(0.4036)
             (  stride 2)   ：
tensor(0.4118)

 
  
  ：
는 2차원 권적과 마찬가지로 권적핵(이 예는 하나의 권적핵이고 차원은 2*3*3*3)을 기억하기만 하면 된다.
 
  
[[[[[ 0.0134,  0.1058, -0.0204],
           [-0.1219,  0.0881, -0.0455],
           [-0.0813,  0.0803,  0.0774]],


          [[-0.0703, -0.0120, -0.0227],
           [-0.0820,  0.0459, -0.0942],
           [ 0.0043, -0.0486, -0.1331]],


          [[-0.1315, -0.0889,  0.1048],
           [-0.1034,  0.0881,  0.0259],
           [-0.0182,  0.0609,  0.0272]]],




         [[[-0.1038,  0.0391,  0.0325],
           [-0.0242,  0.1138,  0.0522],
           [ 0.0649,  0.0600, -0.0476]],


          [[ 0.1099, -0.0557,  0.0714],
           [-0.0997, -0.0301, -0.0354],
           [-0.1291, -0.0227,  0.0915]],


          [[-0.0529, -0.0907,  0.0566],
           [-0.0386, -0.0340, -0.0872],
           [-0.0612, -0.0455, -0.0208]]]]]
 그리고 그 안의 차원과 입력 그림의 차원을 곱하고 마지막에 편중을 더하면 된다.