Hello
I am new to the study of neural networks.
I am trying to improve the CNN based on stanford dogs dataset kaggle dataset, but biggest Accuracy is 12%.
I don’t know what to do, please tell me how improve my CNN, or what to change.
Thanks
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torchvision
from torchvision import datasets, transforms
import os
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 68, 3)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(68, 128, 3)
self.conv3 = nn.Conv2d(128, 128, 3)
self.fc1 = nn.Linear(128 * 26 * 26, 5000)
self.fc2 = nn.Linear(5000, 4000)
self.fc3 = nn.Linear(4000, 120)
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)))
#print(x.shape)
x = x.view(-1, 128 * 26 * 26)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = self.fc3(x)
return x
batch_size=50
epochs=5
lr=0.01
log_interval = 40
classes=os.listdir("/kaggle/input/stanford-dogs-dataset-traintest/cropped/test")
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
torch.manual_seed(1)
transform = transforms.Compose([
transforms.ToTensor(),
#transforms.Normalize((0.1307,), (0.3081,)),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
traindataset = torchvision.datasets.ImageFolder(root='/kaggle/input/stanford-dogs-dataset-traintest/cropped/train', transform=transform)
testdataset = torchvision.datasets.ImageFolder(root='/kaggle/input/stanford-dogs-dataset-traintest/cropped/test', transform=transform)
train_loader = torch.utils.data.DataLoader(traindataset, batch_size=batch_size, shuffle=True, num_workers=1)
test_loader = torch.utils.data.DataLoader(testdataset, batch_size=batch_size, shuffle=False, num_workers=1)
import torchvision.models as models
#model = models.densenet121().to(device)
model = Net().to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=lr, momentum=0.9)
scheduler = torch.optim.lr_scheduler.CyclicLR(optimizer, base_lr=0.01, max_lr=0.1)
#x = next(iter(test_loader))
#model.forward(x[0])
def train(model, train_loader, optimizer, epoch, log_interval):
model.train()
avg_loss = 0
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target .to(device)
optimizer.zero_grad()
output = model(data)
loss = criterion(output, target)
loss.backward()
optimizer.step()
avg_loss+=loss.item()
scheduler.step()
if batch_idx % log_interval == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))
avg_loss/=len(train_loader.dataset)
return avg_loss
def test(model, test_loader, epoch):
model.eval()
test_loss = 0
correct = 0
class_correct = list(0. for i in range(len(classes)))
class_total = list(0. for i in range(len(classes)))
with torch.no_grad():
for data in test_loader:
images, labels = data
images, labels = images.to(device), labels.to(device)
output = model(images)
test_loss += criterion(output, labels).item() # sum up batch loss
_, predicted = torch.max(output, 1)
correct += (predicted == labels).sum().item()
c = (predicted == labels).squeeze()
for i in range(batch_size):
if len(labels) > i:
#print(i)
#print(labels)
label = labels[i]
class_correct[label] += c[i].item()
class_total[label] += 1
test_loss /= len(test_loader.dataset)
for i in range(len(classes)):
if epoch >= 30:
print('Accuracy of %5s : %2d %%' % (classes[i], 100 * class_correct[i] / class_total[i]))
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
test_loss, correct, len(test_loader.dataset),
100. * correct / len(test_loader.dataset)))
accuracy = 100. * correct / len(test_loader.dataset)
return test_loss,accuracy
train_losses = []
test_losses = []
accuracy_list = []
for epoch in range(1, epochs + 1):
trn_loss = train(model, train_loader, optimizer, epoch, log_interval)
test_loss, accuracy = test(model, test_loader, epoch)
train_losses.append(trn_loss)
test_losses.append(test_loss)
accuracy_list.append(accuracy)
Train Epoch: 1 [0/12000 (0%)] Loss: 4.787573
Train Epoch: 1 [2000/12000 (17%)] Loss: 4.757991
Train Epoch: 1 [4000/12000 (33%)] Loss: 4.487531
Train Epoch: 1 [6000/12000 (50%)] Loss: 4.559648
Train Epoch: 1 [8000/12000 (67%)] Loss: 4.353319
Train Epoch: 1 [10000/12000 (83%)] Loss: 4.515636
Test set: Average loss: 0.0860, Accuracy: 345/8580 (4%)
Train Epoch: 2 [0/12000 (0%)] Loss: 4.300442
Train Epoch: 2 [2000/12000 (17%)] Loss: 4.482271
Train Epoch: 2 [4000/12000 (33%)] Loss: 4.314884
Train Epoch: 2 [6000/12000 (50%)] Loss: 4.437295
Train Epoch: 2 [8000/12000 (67%)] Loss: 4.060574
Train Epoch: 2 [10000/12000 (83%)] Loss: 4.061174
Test set: Average loss: 0.0834, Accuracy: 545/8580 (6%)
Train Epoch: 3 [0/12000 (0%)] Loss: 4.054270
Train Epoch: 3 [2000/12000 (17%)] Loss: 4.039870
Train Epoch: 3 [4000/12000 (33%)] Loss: 4.119386
Train Epoch: 3 [6000/12000 (50%)] Loss: 3.935216
Train Epoch: 3 [8000/12000 (67%)] Loss: 4.114314
Train Epoch: 3 [10000/12000 (83%)] Loss: 4.203525
Test set: Average loss: 0.0804, Accuracy: 665/8580 (8%)
Train Epoch: 4 [0/12000 (0%)] Loss: 3.573288
Train Epoch: 4 [2000/12000 (17%)] Loss: 3.669470
Train Epoch: 4 [4000/12000 (33%)] Loss: 3.616935
Train Epoch: 4 [6000/12000 (50%)] Loss: 3.620514
Train Epoch: 4 [8000/12000 (67%)] Loss: 3.837484
Train Epoch: 4 [10000/12000 (83%)] Loss: 3.677999
Test set: Average loss: 0.0792, Accuracy: 787/8580 (9%)
Train Epoch: 5 [0/12000 (0%)] Loss: 3.164417
Train Epoch: 5 [2000/12000 (17%)] Loss: 2.834144
Train Epoch: 5 [4000/12000 (33%)] Loss: 3.117025
Train Epoch: 5 [6000/12000 (50%)] Loss: 3.289179
Train Epoch: 5 [8000/12000 (67%)] Loss: 3.233259
Train Epoch: 5 [10000/12000 (83%)] Loss: 3.406269
Test set: Average loss: 0.0789, Accuracy: 961/8580 (11%)