'모두를 위한 딥러닝 시즌 2' 강의를 듣고 공부하는 스터디 입니다. https://deeplearningzerotoall.github.io/season2/lec_tensorflow.html
비대면 17 May, 2023
10-2 mnist cnn
[총정리]
- 라이브러리 가져오기
# Lab 11 MNIST and Convolutional Neural Network
import torch
import torchvision.datasets as dsets
import torchvision.transforms as transforms
import torch.nn.init
- gpu 사용 설정
device = 'cuda' if torch.cuda.is_available() else 'cpu'
# for reproducibility
torch.manual_seed(777)
if device == 'cuda':
torch.cuda.manual_seed_all(777)
- 학습에 사용되는 parameter 설정
# parameters
learning_rate = 0.001
training_epochs = 15
batch_size = 100
- 데이터셋 가져오고 loader 만들기
# MNIST dataset
mnist_train = dsets.MNIST(root='MNIST_data/',
train=True,
transform=transforms.ToTensor(),
download=True)
mnist_test = dsets.MNIST(root='MNIST_data/',
train=False,
transform=transforms.ToTensor(),
download=True)
# dataset loader
data_loader = torch.utils.data.DataLoader(dataset=mnist_train,
batch_size=batch_size,
shuffle=True,
drop_last=True)
- 학습 모델 만들기(torch.nn.module)
# CNN Model (2 conv layers)
class CNN(torch.nn.Module):
def __init__(self):
super(CNN, self).__init__()
# L1 ImgIn shape=(?, 28, 28, 1)
# Conv -> (?, 28, 28, 32)
# Pool -> (?, 14, 14, 32)
self.layer1 = torch.nn.Sequential(
torch.nn.Conv2d(1, 32, kernel_size=3, stride=1, padding=1),
torch.nn.ReLU(),
torch.nn.MaxPool2d(kernel_size=2, stride=2))
# L2 ImgIn shape=(?, 14, 14, 32)
# Conv ->(?, 14, 14, 64)
# Pool ->(?, 7, 7, 64)
self.layer2 = torch.nn.Sequential(
torch.nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1),
torch.nn.ReLU(),
torch.nn.MaxPool2d(kernel_size=2, stride=2))
# Final FC 7x7x64 inputs -> 10 outputs(0~9)
self.fc = torch.nn.Linear(7 * 7 * 64, 10, bias=True) #fully connected layer
torch.nn.init.xavier_uniform_(self.fc.weight) #weight 초기화
def forward(self, x):
out = self.layer1(x) #layer1 통과
#print(out.shape)로 확인
out = self.layer2(out) #layer2 통과
out = out.view(out.size(0), -1) # Flatten them for FC
out = self.fc(out)
return out
# instantiate CNN model
model = CNN().to(device)
- loss function(criterion) 선택, 최적화 도구(optimizer) 선택
# define cost/loss & optimizer
criterion = torch.nn.CrossEntropyLoss().to(device) # Softmax is internally computed.
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
- 모델 학습 및 loss check
# train my model
total_batch = len(data_loader)
print('Learning started. It takes sometime.')
for epoch in range(training_epochs):
avg_cost = 0
for X, Y in data_loader:
# image is already size of (28x28), no reshape
# label is not one-hot encoded
X = X.to(device) #torch cuda 연산 진행
Y = Y.to(device)
optimizer.zero_grad()
hypothesis = model(X)
cost = criterion(hypothesis, Y)
cost.backward()
optimizer.step()
avg_cost += cost / total_batch
print('[Epoch: {:>4}] cost = {:>.9}'.format(epoch + 1, avg_cost))
print('Learning Finished!')
- 학습된 모델 성능 확인
# Test model and check accuracy
with torch.no_grad():
X_test = mnist_test.test_data.view(len(mnist_test), 1, 28, 28).float().to(device)
Y_test = mnist_test.test_labels.to(device)
prediction = model(X_test)
correct_prediction = torch.argmax(prediction, 1) == Y_test
accuracy = correct_prediction.float().mean()
print('Accuracy:', accuracy.item())
더 많은 layer의 CNN
- 학습 모델 만들기(torch.nn.module)
# CNN Model
class CNN(torch.nn.Module):
def __init__(self):
super(CNN, self).__init__()
self.keep_prob = 0.5
# L1 ImgIn shape=(?, 28, 28, 1)
# Conv -> (?, 28, 28, 32)
# Pool -> (?, 14, 14, 32)
self.layer1 = torch.nn.Sequential(
torch.nn.Conv2d(1, 32, kernel_size=3, stride=1, padding=1),
torch.nn.ReLU(),
torch.nn.MaxPool2d(kernel_size=2, stride=2))
# L2 ImgIn shape=(?, 14, 14, 32)
# Conv ->(?, 14, 14, 64)
# Pool ->(?, 7, 7, 64)
self.layer2 = torch.nn.Sequential(
torch.nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1),
torch.nn.ReLU(),
torch.nn.MaxPool2d(kernel_size=2, stride=2))
# L3 ImgIn shape=(?, 7, 7, 64)
# Conv ->(?, 7, 7, 128)
# Pool ->(?, 4, 4, 128)
self.layer3 = torch.nn.Sequential(
torch.nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1),
torch.nn.ReLU(),
torch.nn.MaxPool2d(kernel_size=2, stride=2, padding=1))
# L4 FC 4x4x128 inputs -> 625 outputs
self.fc1 = torch.nn.Linear(4 * 4 * 128, 625, bias=True)
torch.nn.init.xavier_uniform_(self.fc1.weight)
self.layer4 = torch.nn.Sequential(
self.fc1,
torch.nn.ReLU(),
torch.nn.Dropout(p=1 - self.keep_prob))
# L5 Final FC 625 inputs -> 10 outputs
self.fc2 = torch.nn.Linear(625, 10, bias=True)
torch.nn.init.xavier_uniform_(self.fc2.weight)
def forward(self, x):
out = self.layer1(x)
out = self.layer2(out)
out = self.layer3(out)
out = out.view(out.size(0), -1) # Flatten them for FC
out = self.layer4(out)
out = self.fc2(out)
#모델에 임의의 값을 넣어 test
value = torch.Tensor[1,1,28,28]
print(model(value).shape)
⇒ 더 깊어져도 모델의 성능이 떨어질 수 있음
⇒ 효율적으로 만드는 것이 중요!
10-3 visdom
-visdom 설치
pip install visdom
-visdom 서버 켜기
python -m visdom.server
visdom 사용법
**import** torch
**import** torch.nn **as** nn
**import** torchvision
**import** torchvision.datasets **as** dsets
#Jupyter Notebook > Terminal 를 새로 켜서 python -m visdom.server 를 입력
#import visdom
**import** visdom
vis **=** visdom**.**Visdom()
#Text
vis.text("Hello, world!",env="main")
#image
a=torch.randn(3,200,200)
vis.image(a)
#images
vis.images(torch.Tensor(3,3,28,28))
##example (using MNIST and CIFAR10)
# 시간이 좀 걸립니다.
MNIST = dsets.MNIST(root="./MNIST_data",train = True,transform=torchvision.transforms.ToTensor(), download=True)
cifar10 = dsets.CIFAR10(root="./cifar10",train = True, transform=torchvision.transforms.ToTensor(),download=True)
CIFAR10
data = cifar10.__getitem__(0)
print(data[0].shape)
vis.images(data[0],env="main")
##MNIST
data = MNIST.__getitem__(0)
print(data[0].shape)
vis.images(data[0],env="main")
##Check dataset
data_loader = torch.utils.data.DataLoader(dataset = MNIST,
batch_size = 32,
shuffle = False)
for num, value in enumerate(data_loader):
value = value[0]
print(value.shape)
vis.images(value)
break
vis.close(env="main")
##Line Plot
Y_data = torch.randn(5)
plt = vis.line (Y=Y_data)
X_data = torch.Tensor([1,2,3,4,5])
plt = vis.line(Y=Y_data, X=X_data)
##Line update
Y_append = torch.randn(1)
X_append = torch.Tensor([6])
vis.line(Y=Y_append, X=X_append, win=plt, update='append')
multiple Line on single windows
num = torch.Tensor(list(range(0,10)))
num = num.view(-1,1)
num = torch.cat((num,num),dim=1)
plt = vis.line(Y=torch.randn(10,2), X = num)
##Line info
plt = vis.line(Y=Y_data, X=X_data, opts = dict(title='Test', showlegend=True))
plt = vis.line(Y=Y_data, X=X_data, opts = dict(title='Test', legend = ['1번'],showlegend=True))
plt = vis.line(Y=torch.randn(10,2), X = num, opts=dict(title='Test', legend=['1번','2번'],showlegend=True))
##make function for update line
def loss_tracker(loss_plot, loss_value, num):
'''num, loss_value, are Tensor'''
vis.line(X=num,
Y=loss_value,
win = loss_plot,
update='append'
)
plt = vis.line(Y=torch.Tensor(1).zero_())
for i in range(500):
loss = torch.randn(1) + i
loss_tracker(plt, loss, torch.Tensor([i]))
##close the window
vis**.**close(env**=**"main")
⇒ visdom 왜 씀? → 모델 학습을 시각적으로 확인할 수 있도록 돕는 tool
cf) tensorboardx도 visualization tool로 많이 쓰임
minist-cnn with visdom
##make plot
loss_plt = vis.line(Y=torch.Tensor(1).zero_(),opts=dict(title='loss_tracker', legend=['loss'], showlegend=True))
##train with loss_tracker
total_batch = len(data_loader)
for epoch in range(training_epochs):
avg_cost = 0
for X, Y in data_loader:
X = X.to(device)
Y = Y.to(device)
optimizer.zero_grad()
hypothesis = model(X)
cost = criterion(hypothesis, Y)
cost.backward()
optimizer.step()
avg_cost += cost / total_batch
print('[Epoch:{}] cost = {}'.format(epoch+1, avg_cost))
**loss_tracker(loss_plt, torch.Tensor([avg_cost]), torch.Tensor([epoch]))**
print('Learning Finished!')
10-4 ImageFolder
? 내가 가진 사진 활용해서 딥러닝 모델 학습하기
- 나만의 데이터 셋 준비하기
-class마다 다르게 폴더를 만들어 이미지들 넣어놓기
-사진 크기 줄이기
#이미지 보이게 할 때 활용
from matplotlib.pyplot import imshow
%matplotlib inline
- torchvision.datasets.ImageFolder으로 불러오기
trans = transforms.Compose([
transforms.Resize((64,128))
])
train_data = torchvision.datasets.ImageFolder(root='custom_data/origin_data', transform=trans)
- transforms 적용하여 저장 하기 origin_data -> train_data
for num, value in enumerate(train_data):
data, label = value
print(num, data, label)
if(label == 0):
data.save('custom_data/train_data/gray/%d_%d.jpeg'%(num, label)) #경로대로 폴더 만들어져 있어야 함
else:
data.save('custom_data/train_data/red/%d_%d.jpeg'%(num, label)
- CNN모델 생성 및 학습
#모델 저장하고 불러오기
torch.save(net.state_dict(), "./model/model.pth")
new_net = CNN().to(device)
new_net.load_state_dict(torch.load('./model/model.pth'))
Q. 왜 저장해놔야 함? 어디에 저장되는 거임?
#test
trans=torchvision.transforms.Compose([
transforms.Resize((64,128)),
transforms.ToTensor()
])
test_data = torchvision.datasets.ImageFolder(root='./custom_data/test_data', transform=trans)
test_set = DataLoader(dataset = test_data, batch_size = len(test_data))
with torch.no_grad():
for num, data in enumerate(test_set):
imgs, label = data
imgs = imgs.to(device)
label = label.to(device)
prediction = net(imgs)
correct_prediction = torch.argmax(prediction, 1) == label
accuracy = correct_prediction.float().mean()
print('Accuracy:', accuracy.item())
대면 20 May, 2023
(sum up)mnist cnn
CIFAR-10 이미지 크기에 맞춰 CNN2 다시 설계
3x32x32
self.layer1 = torch.nn.Sequential(
nn.Conv2d(3,x,kernel_size=3, stride=1, padding=0),
nn.ReLU(),
nn.MaxPool2d(2),
)
self.layer2 = nn.Sequential(
nn.Conv2d(x,y,kernel_size=3, stride=1, padding=0),
nn.ReLU(),
nn.MaxPool2d(2),
)
self.layer3 = nn.Sequential(
nn.Linear(inputsize_height->(outputsize 식/maxpool1/maxpool2)*inputsize_width->(outputsize 식/maxpool1/maxpool2)*y, 120),
nn.ReLU(),
nn.Linear(120,10)
)
Top comments (0)