In this notebook we'll cover some of the basics for 2-d convolution for image classification using the MNIST dataset.
import torch
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
# device configuration
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
num_epochs = 5
num_classes = 10
learning_rate = 0.001
train_dataset = torchvision.datasets.MNIST(root='./MNIST',
train=True,
transform=transforms.ToTensor(),
download=True)
test_dataset = torchvision.datasets.MNIST(root='./MNIST',
train=False,
transform=transforms.ToTensor())
batch_size = 100 # how many examples are processed at each step
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=batch_size,
shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=batch_size,
shuffle=False)
class ConvNet1(nn.Module):
def __init__(self, num_classes=10):
super(ConvNet1, self).__init__()
self.layer1 = nn.Sequential(
nn.Conv2d(1, 16, kernel_size=5, stride=1, padding=2),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.layer2 = nn.Sequential(
nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=2),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.fc = nn.Linear(7*7*32, num_classes)
def forward(self, x):
out = self.layer1(x)
out = self.layer2(out)
out = out.reshape(out.size(0), -1)
out = self.fc(out)
return out
model = ConvNet1(num_classes).to(device)
# Loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
num_batches = len(train_loader)
for epoch in range(num_epochs):
for i, (images, labels) in enumerate(train_loader):
images = images.to(device)
labels = labels.to(device)
# forward pass
outputs = model(images)
loss = criterion(outputs, labels)
# backward and optimize
optimizer.zero_grad()
loss.backward()
optimizer.step()
if (i+1) % 100 == 0:
print ('epoch [{}/{}], batch [{}/{}], loss: {:.4f}'
.format(epoch+1, num_epochs, i+1, num_batches, loss.item()))
epoch [1/5], batch [100/600], loss: 0.3148 epoch [1/5], batch [200/600], loss: 0.1325 epoch [1/5], batch [300/600], loss: 0.2210 epoch [1/5], batch [400/600], loss: 0.0793 epoch [1/5], batch [500/600], loss: 0.2944 epoch [1/5], batch [600/600], loss: 0.1435 epoch [2/5], batch [100/600], loss: 0.0370 epoch [2/5], batch [200/600], loss: 0.0773 epoch [2/5], batch [300/600], loss: 0.0421 epoch [2/5], batch [400/600], loss: 0.0519 epoch [2/5], batch [500/600], loss: 0.0405 epoch [2/5], batch [600/600], loss: 0.0763 epoch [3/5], batch [100/600], loss: 0.0544 epoch [3/5], batch [200/600], loss: 0.0341 epoch [3/5], batch [300/600], loss: 0.1854 epoch [3/5], batch [400/600], loss: 0.0307 epoch [3/5], batch [500/600], loss: 0.0335 epoch [3/5], batch [600/600], loss: 0.0378 epoch [4/5], batch [100/600], loss: 0.1171 epoch [4/5], batch [200/600], loss: 0.0608 epoch [4/5], batch [300/600], loss: 0.0269 epoch [4/5], batch [400/600], loss: 0.0056 epoch [4/5], batch [500/600], loss: 0.0202 epoch [4/5], batch [600/600], loss: 0.0112 epoch [5/5], batch [100/600], loss: 0.0295 epoch [5/5], batch [200/600], loss: 0.0119 epoch [5/5], batch [300/600], loss: 0.0657 epoch [5/5], batch [400/600], loss: 0.0099 epoch [5/5], batch [500/600], loss: 0.0647 epoch [5/5], batch [600/600], loss: 0.0025
# test the model
model.eval() # eval mode (batchnorm uses moving mean/variance instead of mini-batch mean/variance)
with torch.no_grad():
correct = 0
total = 0
for images, labels in test_loader:
images = images.to(device)
labels = labels.to(device)
outputs = model(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Test Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total))
ConvNet1( (layer1): Sequential( (0): Conv2d(1, 16, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (1): ReLU() (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (layer2): Sequential( (0): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (1): ReLU() (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (fc): Linear(in_features=1568, out_features=10, bias=True) )
Test Accuracy of the model on the 10000 test images: 98.88 %
class ConvNet2(nn.Module):
def __init__(self, num_classes=10):
super(ConvNet2, self).__init__()
self.layer1 = nn.Sequential(
nn.Conv2d(1, 16, kernel_size=5, stride=1, padding=2),
nn.BatchNorm2d(16),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.layer2 = nn.Sequential(
nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=2),
nn.BatchNorm2d(32),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.drop_out = nn.Dropout()
self.fc = nn.Linear(7*7*32, num_classes)
def forward(self, x):
out = self.layer1(x)
out = self.layer2(out)
out = out.reshape(out.size(0), -1) # flatten
out = self.drop_out(out)
out = self.fc(out)
return out
model = ConvNet2(num_classes).to(device)
# Loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
num_batches = len(train_loader)
for epoch in range(num_epochs):
for i, (images, labels) in enumerate(train_loader):
images = images.to(device)
labels = labels.to(device)
# forward pass
outputs = model(images)
loss = criterion(outputs, labels)
# backward and optimize
optimizer.zero_grad()
loss.backward()
optimizer.step()
if (i+1) % 100 == 0:
print ('epoch [{}/{}], batch [{}/{}], loss: {:.4f}'
.format(epoch+1, num_epochs, i+1, num_batches, loss.item()))
epoch [1/5], batch [100/600], loss: 0.3027 epoch [1/5], batch [200/600], loss: 0.2901 epoch [1/5], batch [300/600], loss: 0.1391 epoch [1/5], batch [400/600], loss: 0.0396 epoch [1/5], batch [500/600], loss: 0.1175 epoch [1/5], batch [600/600], loss: 0.0288 epoch [2/5], batch [100/600], loss: 0.0374 epoch [2/5], batch [200/600], loss: 0.0803 epoch [2/5], batch [300/600], loss: 0.0305 epoch [2/5], batch [400/600], loss: 0.0406 epoch [2/5], batch [500/600], loss: 0.0463 epoch [2/5], batch [600/600], loss: 0.0810 epoch [3/5], batch [100/600], loss: 0.0653 epoch [3/5], batch [200/600], loss: 0.0808 epoch [3/5], batch [300/600], loss: 0.0590 epoch [3/5], batch [400/600], loss: 0.0286 epoch [3/5], batch [500/600], loss: 0.0262 epoch [3/5], batch [600/600], loss: 0.0597 epoch [4/5], batch [100/600], loss: 0.0544 epoch [4/5], batch [200/600], loss: 0.0353 epoch [4/5], batch [300/600], loss: 0.0306 epoch [4/5], batch [400/600], loss: 0.0238 epoch [4/5], batch [500/600], loss: 0.0596 epoch [4/5], batch [600/600], loss: 0.0065 epoch [5/5], batch [100/600], loss: 0.0402 epoch [5/5], batch [200/600], loss: 0.0399 epoch [5/5], batch [300/600], loss: 0.0305 epoch [5/5], batch [400/600], loss: 0.0487 epoch [5/5], batch [500/600], loss: 0.0386 epoch [5/5], batch [600/600], loss: 0.0382
# test the model
model.eval() # eval mode (batchnorm uses moving mean/variance instead of mini-batch mean/variance)
with torch.no_grad():
correct = 0
total = 0
for images, labels in test_loader:
images = images.to(device)
labels = labels.to(device)
outputs = model(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Test Accuracy of the model on the 10000 test images: {} %'.format(100 * correct / total))
ConvNet2( (layer1): Sequential( (0): Conv2d(1, 16, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (2): ReLU() (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (layer2): Sequential( (0): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2)) (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (2): ReLU() (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (drop_out): Dropout(p=0.5) (fc): Linear(in_features=1568, out_features=10, bias=True) )
Test Accuracy of the model on the 10000 test images: 98.95 %