24 lines
776 B
Python
24 lines
776 B
Python
|
import torch.nn as nn
|
||
|
import torch.nn.functional as F
|
||
|
|
||
|
class CNNModel(nn.Module): #model of the CNN type
|
||
|
def __init__(self):
|
||
|
super(CNNModel, self).__init__()
|
||
|
self.conv1 = nn.Conv2d(3, 32, 5)
|
||
|
self.conv2 = nn.Conv2d(32, 64, 5)
|
||
|
self.conv3 = nn.Conv2d(64, 128, 3)
|
||
|
self.conv4 = nn.Conv2d(128, 256, 5)
|
||
|
|
||
|
self.fc1 = nn.Linear(256, 50)
|
||
|
|
||
|
self.pool = nn.MaxPool2d(2, 2)
|
||
|
|
||
|
def forward(self, x):
|
||
|
x = self.pool(F.relu(self.conv1(x)))
|
||
|
x = self.pool(F.relu(self.conv2(x)))
|
||
|
x = self.pool(F.relu(self.conv3(x)))
|
||
|
x = self.pool(F.relu(self.conv4(x)))
|
||
|
bs, _, _, _ = x.shape
|
||
|
x = F.adaptive_avg_pool2d(x, 1).reshape(bs, -1)
|
||
|
x = self.fc1(x)
|
||
|
return x
|