When i’m using pytorch resnet model, model not converging,but when using other architecture model the training process work will.
configurations :
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
Resnet model
model_ft = models.resnet18(pretrained=True)
num_ftrs = model_ft.fc.in_features
model_ft.fc = nn.Linear(num_ftrs, 10)
net = model_ft
training steps:
[1, 4] loss: 526.469 acc: 348.250
1393
[2, 4] loss: 529.261 acc: 346.750
1387
[3, 4] loss: 532.855 acc: 349.250
1397
[4, 4] loss: 530.508 acc: 348.000
1392
[5, 4] loss: 526.442 acc: 347.000
1388
[6, 4] loss: 531.521 acc: 348.500
1394
[7, 4] loss: 527.574 acc: 348.250
1393
[8, 4] loss: 527.463 acc: 349.000
1396
[9, 4] loss: 527.096 acc: 347.250
1389
[10, 4] loss: 529.270 acc: 348.750
1395
[11, 4] loss: 527.993 acc: 348.500
1394
[12, 4] loss: 532.372 acc: 349.000
1396
[13, 4] loss: 530.069 acc: 348.250
1393
[14, 4] loss: 531.173 acc: 348.000
1392
[15, 4] loss: 530.445 acc: 347.250
1389
Finished Training
type or paste code here
example for model that works well
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 6, 5)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16 * 13 * 13, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)
def forward(self, x):
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = x.view(-1, 16 * 13 * 13)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
training steps:
[1, 4] loss: 578.536 acc: 355.250
1421
[2, 4] loss: 432.506 acc: 438.250
1753
[3, 4] loss: 359.835 acc: 503.250
2013
[4, 4] loss: 289.072 acc: 544.250
2177
[5, 4] loss: 251.283 acc: 568.500
2274
[6, 4] loss: 237.213 acc: 575.750
2303
[7, 4] loss: 207.391 acc: 590.250
2361
[8, 4] loss: 192.430 acc: 598.500
2394
[9, 4] loss: 178.400 acc: 605.250
2421
[10, 4] loss: 172.829 acc: 609.500
2438
[11, 4] loss: 150.899 acc: 617.000
2468
[12, 4] loss: 160.563 acc: 614.750
2459
[13, 4] loss: 139.884 acc: 623.750
2495
[14, 4] loss: 124.817 acc: 629.000
2516
[15, 4] loss: 132.767 acc: 626.000
2504
Finished Training