PyTorch를 활용한 전이 학습

import pandas as pd
import os
import torch
import torchvision
import torchvision.transforms as transforms
from torch.utils.data import DataLoader, random_split
from torchvision import models
from torchvision import datasets, transforms
import torch.nn as nn
import torch.optim as optim
import zipfile


!pip install wandb
import wandb
os.environ['CUDA_LAUNCH_BLOCKING'] = "1"

from google.colab import drive
drive.mount('/content/drive')


!pip install kaggle
os.environ['KAGGLE_CONFIG_DIR'] = '/content/drive/MyDrive/kaggle'
!kaggle datasets download -d imbikramsaha/caltech-101


file_path = '/content/caltech-101.zip' 


with zipfile.ZipFile(file_path, 'r') as zip_ref:
    zip_ref.extractall('/content/kaggle/') 

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])


dataset = datasets.ImageFolder(root='/content/kaggle/caltech-101', transform=transform)


total_size = len(dataset)
train_size = int(0.8 * total_size) 
test_size = total_size - train_size


train_dataset, test_dataset = random_split(dataset, [train_size, test_size])


train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=32, shuffle=False)

wandb.init(project="transfer_learn_pytorch", name="caltech101_classification_2")

vgg = models.vgg16(pretrained=True)


for param in vgg.features.parameters():
    param.requires_grad = False


num_features = vgg.classifier[6].in_features
vgg.classifier[6] = nn.Linear(num_features, 102)  # Change the last layer


device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(device)
vgg.to(device)

config = wandb.config
config.learning_rate = 0.001
config.momentum = 0.9
config.epochs = 5
config.batch_size = 32

criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(vgg.classifier.parameters(), lr=0.001, momentum=0.9)


def train_model(model, criterion, optimizer, num_epochs):
    for epoch in range(num_epochs):
        model.train()
        running_loss = 0.0
        for inputs, labels in train_loader:
            inputs, labels = inputs.to(device), labels.to(device)


            optimizer.zero_grad()
            outputs = model(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()


            running_loss += loss.item()


        epoch_loss = running_loss / len(train_loader)
        print(f"Epoch {epoch+1}, Loss: {epoch_loss}")


        wandb.log({"epoch": epoch + 1, "loss": epoch_loss})


train_model(vgg, criterion, optimizer, num_epochs=5)

def evaluate_model(model, criterion, loader):
    model.eval()  
    total_loss = 0.0
    total = 0


    with torch.no_grad(): 
        for inputs, labels in loader:
            inputs, labels = inputs.to(device), labels.to(device)
            outputs = model(inputs)
            loss = criterion(outputs, labels)
            total_loss += loss.item() * inputs.size(0)
            total += inputs.size(0)


    average_loss = total_loss / total
    return average_loss

pretrained_vgg = models.vgg16(pretrained=True).to(device)


pretrained_test_loss = evaluate_model(pretrained_vgg, criterion, test_loader)


wandb.log({"pretrained_test_loss": pretrained_test_loss})

test_loss = evaluate_model(vgg, criterion, test_loader)


wandb.log({"fine_tuned_test_loss": test_loss})

def log_predictions(model, loader, prefix=""):
    model.eval()
    with torch.no_grad():
        for inputs, labels in loader:
            inputs, labels = inputs.to(device), labels.to(device)
            outputs = model(inputs)
            _, preds = torch.max(outputs, 1)


            wandb.log({f"{prefix}examples": [wandb.Image(x.cpu(), caption=f"Pred:{pred.item()}, Label:{y.item()}")
                                    for x, pred, y in zip(inputs, preds, labels)]})
            break 

log_predictions(pretrained_vgg, test_loader, prefix="pretrained_")

log_predictions(vgg, test_loader, prefix="fine_tuned_")