RuntimeError: mat1 and mat2 shapes cannot be multiplied (4x73034 and 200x120)

Question

为皮肤检测数据集构建神经网络层，并在此处出错。我知道我犯了一些错误，但无法弄清楚。获取图像大小 224*224 和通道 3 后出现错误：RuntimeError: mat1 和 mat2 形状无法相乘（4x73034 和 200x120）

import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):
  def __init__(self):
    super(Net, self).__init__()
    self.conv1 = nn.Conv2d(3, 16, 5)
    self.pool = nn.MaxPool2d(2, 2)
    self.conv2 = nn.Conv2d(16, 26, 5)
    self.fc1 = nn.Linear(8 * 5 * 5, 120)
    self.fc2 = nn.Linear(120, 86)
    self.fc3 = nn.Linear(86, 2)

  def forward(self, x):
    x = self.pool(F.relu(self.conv1(x)))
    x = self.pool(F.relu(self.conv2(x)))
    x = torch.flatten(x,1)
    x = F.relu(self.fc1(x))
    x = F.relu(self.fc2(x))
    x = self.fc3(x)
    return x

net = Net().to(device)
print(net)

这些是图层和网络模块

<ipython-input-41-8c9bafb31c44> in forward(self, x)
     16     x = self.pool(F.relu(self.conv2(x)))
     17     x = torch.flatten(x,1)
---> 18     x = F.relu(self.fc1(x))
     19     x = F.relu(self.fc2(x))
     20     x = self.fc3(x)

谁能帮我解决这个问题。

Answer 1

torch.flatten 的输出与 self.fc1

的输入不匹配

打印torch.flatten

输出的形状

x = torch.flatten(x,1)
print(x.size())

并随后更新 self.fc1

的定义

self.fc1 = nn.Linear(8 * 5 * 5, 120)

Answer 2

正如Anant所说，你需要匹配扁平化的conv2维度（73034）作为fc1层的输入维度。

self.fc1 = nn.Linear(73034, 120)

计算每个conv层输出的公式：

[(height or width) - kernel size + 2*padding] / stride + 1

对于以下内容，我将使用尺寸（通道、高度、宽度）输入 (3,224,224) -> conv1 -> (16,220,220) -> pool -> (16,110,110) -> conv2 -> (26,106,106) -> pool -> (26,53,53) -> flatten -> (73034)

你的batch size好像是4，指的是(4x73034)中的“4”。如果打印 conv1 或 conv2 层的输出尺寸，格式将为 (Batch, Channels, Height, Width)。

RuntimeError: mat1 and mat2 shapes cannot be multiplied (4x73034 and 200x120)

RuntimeError: mat1 and mat2 shapes cannot be multiplied (4x73034 and 200x120)

python

machine-learning

deep-learning

pytorch

conv-neural-network