라이브러리 import, CIFAR10 다운
from __future__ import print_function
import os
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])
train_set = torchvision.datasets.CIFAR10('./datasets', train=True,
download=True, transform=transform)
test_set = torchvision.datasets.CIFAR10('./datasets', train=False,
download=True, transform=transform)
train_loader = torch.utils.data.DataLoader(train_set, batch_size=128,
shuffle=True, num_workers=4)
test_loader = torch.utils.data.DataLoader(test_set, batch_size=100,
shuffle=False, num_workers=4)
classes = ('plane', 'car', 'bird', 'cat', 'deer',
'dog', 'frog', 'horse', 'ship', 'truck')
VGG 모델 구현
cfg = {
'VGG11': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'VGG13': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'VGG16': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
'VGG19': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
}
class VGG(nn.Module):
def __init__(self, vgg_name):
super(VGG, self).__init__()
self.features = self._make_layers(cfg[vgg_name])
self.classifier = nn.Linear(512, 10)
self._initialize_weight()
def forward(self, x):
out = self.features(x)
out = out.view(out.size(0), -1)
out = self.classifier(out)
return out
def _make_layers(self, cfg):
layers = []
in_channels = 3
for x in cfg:
if x == 'M':
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
else:
layers += [nn.Conv2d(in_channels, x, kernel_size=3, padding=1),
nn.BatchNorm2d(x),
nn.ReLU(inplace=True)]
in_channels = x
layers += [nn.AvgPool2d(kernel_size=1, stride=1)]
return nn.Sequential(*layers)
def _initialize_weight(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
# xavier is used in VGG's paper
nn.init.xavier_normal_(m.weight.data)
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
elif isinstance(m, nn.Linear):
m.weight.data.normal_(0, 0.01)
m.bias.data.zero_()
model = VGG('VGG16').to(device)
print(model)
if device == 'cuda':
model = nn.DataParallel(model)
torch.backends.cudnn.benchmark = TrueVGG 16 으로 training
Training, Test 과정 수정 전 Acc : 90%
# Training
def train(epoch):
print('\nEpoch: %d' % (epoch + 1))
net.train()
train_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(train_loader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
outputs = net(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
train_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
if batch_idx % 100 == 99: # print every 100 mini-batches
print('[%d, %5d] loss: %.5f | Acc: %.3f%% (%d/%d)' %
(epoch + 1, batch_idx + 1, train_loss / 2000, 100.*correct/total, correct, total))
train_loss = 0.0
total = 0
correct = 0
load_model = True
start_epoch = 0
print('start_epoch: %s' % start_epoch)correct = 0
total = 0
with torch.no_grad():
for data in test_loader:
images, labels = data
images, labels = images.to(device), labels.to(device)
outputs = net(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Accuracy of the network on the 10000 test images: %d %%' % (
100 * correct / total))training 진행 후 test data로 acc 측정
Training, Test 과정 수정 후 Acc : 92%
for epoch in range(num_epoch):
print(f"====== { epoch+1} epoch of { num_epoch } ======")
model.train()
#scheduler2.step()
lr_scheduler(optimizer, epoch)
train_loss = 0
valid_loss = 0
correct = 0
total_cnt = 0
# Train Phase
for step, batch in enumerate(train_loader):
# input and target
batch[0], batch[1] = batch[0].to(device), batch[1].to(device)
optimizer.zero_grad()
logits = model(batch[0])
loss = loss_fn(logits, batch[1])
loss.backward()
optimizer.step()
train_loss += loss.item()
_, predict = logits.max(1)
total_cnt += batch[1].size(0)
correct += predict.eq(batch[1]).sum().item()
if step % 100 == 0 and step != 0:
print(f"\n====== { step } Step of { len(train_loader) } ======")
print(f"Train Acc : { correct / total_cnt }")
print(f"Train Loss : { loss.item() / batch[1].size(0) }")
correct = 0
total_cnt = 0
# Test Phase
with torch.no_grad():
model.eval()
for step, batch in enumerate(test_loader):
# input and target
batch[0], batch[1] = batch[0].to(device), batch[1].to(device)
total_cnt += batch[1].size(0)
logits = model(batch[0])
valid_loss += loss_fn(logits, batch[1])
_, predict = logits.max(1)
correct += predict.eq(batch[1]).sum().item()
valid_acc = correct / total_cnt
print(f"\nValid Acc : { valid_acc }")
print(f"Valid Loss : { valid_loss / total_cnt }")
if(valid_acc > best_acc):
best_acc = valid_acc
torch.save(model, model_name)
print("Model Saved!")
if epoch % 1 == 0: # 매 10 iteration마다 업데이트
# writer.add_scalar('train_loss', loss.item() / batch[1].size(0), epoch)
writer.add_scalar('test_loss', valid_loss / total_cnt, epoch)
writer.add_scalar('train_acc', correct / total_cnt, epoch)
writer.add_scalar('test_acc', valid_acc, epoch)
writer.closebest acc 일때 model save
