用pytorch实现预训练网络的finetune

继续熟悉pytorch，发现的确比TensorFlow好用，在官网finetune教程的基础上进行了大幅修改，主要是熟悉了pytorch自带的Dataset和DataLoader类。

# -*- coding: utf-8 -*-

import os, torch, glob, time, copy
import numpy as np
from torch.autograd import Variable
from PIL import Image  
from torchvision import models, transforms
import torch.nn as nn
from torch.utils.data import DataLoader, Dataset
import torch.optim as optim
from torch.optim import lr_scheduler

train_data_dir = './hymenoptera_data/train'
test_data_dir = './hymenoptera_data/val'

def get_lists(data_dir):        
    files_list = []
    labels_list = []
    sub_dirs = [x[0] for x in os.walk(data_dir) ]
    sub_dirs = sub_dirs[1:]
    
    extentions = ['jpg', 'jpeg', 'JPG', 'JPEG']
        
    for sub_dir in sub_dirs:
        if os.path.basename(sub_dir) == 'ants':
            label = [0]
        elif os.path.basename(sub_dir) == 'bees':
            label = [1] 
            
        for extention in extentions:
            file_glob = os.path.join(sub_dir, '*.' + extention)
            new_files = glob.glob(file_glob)
            files_list.extend(new_files)
            labels_list = labels_list + label * len(new_files)
    return files_list, labels_list

class myDataset(Dataset):
    # TensorDataset继承Dataset, 重载了__init__, __getitem__, __len__
    # 实现将一组Tensor数据对封装成Tensor数据集
    # 能够通过index得到数据集的数据，能够通过len，得到数据集大小

    def __init__(self, files_list, labels_list, transform):
        self.files_list = files_list
        self.labels_list = labels_list
        self.transform = transform

    def __getitem__(self, index):
        img = Image.open(self.files_list[index])
        img = self.transform(img)
        return img, self.labels_list[index]

    def __len__(self):
        return len(self.labels_list)
    
def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
    since1 = time.time()

    best_model_wts = copy.deepcopy(model.state_dict())
    best_acc = 0.0

    for epoch in range(num_epochs):
        since2 = time.time()
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        print('-' * 10)

        # Each epoch has a training and validation phase
        for phase in ['train', 'val']:
            if phase == 'train':
                scheduler.step()
                model.train(True)  # Set model to training mode
            else:
                model.train(False)  # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0

            # Iterate over data.
            for data in (dataloaders[phase]):
                # get the inputs
                inputs, labels = data

                # wrap them in Variable
                if use_gpu:
                    inputs = Variable(inputs.cuda())
                    labels = Variable(labels.cuda())
                else:
                    inputs, labels = Variable(inputs), Variable(labels)

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                outputs = model(inputs)
                _, preds = torch.max(outputs.data, 1)
                loss = criterion(outputs, labels)

                # backward + optimize only if in training phase
                if phase == 'train':
                    loss.backward()
                    optimizer.step()

                # statistics
                running_loss += loss.data[0] * inputs.size(0)
                running_corrects += torch.sum(preds == labels.data)

            epoch_loss = running_loss / dataset_sizes[phase]
            epoch_acc = running_corrects / dataset_sizes[phase]

            print('{} Loss: {:.4f} Acc: {:.4f}'.format(
                phase, epoch_loss, epoch_acc))

            

            # deep copy the model
            if phase == 'val' and epoch_acc > best_acc:
                best_acc = epoch_acc
                best_model_wts = copy.deepcopy(model.state_dict())
                
        print(time.time() - since2)


    time_elapsed = time.time() - since1
    print('Training complete in {:.0f}m {:.0f}s'.format(
        time_elapsed // 60, time_elapsed % 60))
    print('Best val Acc: {:4f}'.format(best_acc))

    # load best model weights
    model.load_state_dict(best_model_wts)
    return model

if __name__ == '__main__':
    BATCH_SIZE = 4
    use_gpu = torch.cuda.is_available()

    transform = transforms.Compose([
            transforms.Scale(256),
            transforms.CenterCrop(224),
            transforms.ToTensor()    ]
    )
    train_images, train_labels = get_lists(train_data_dir)
    test_images, test_labels = get_lists(test_data_dir)

    train_dataset = myDataset(train_images, train_labels, transform)
    test_dataset = myDataset(test_images, test_labels, transform)
    train_loader = DataLoader(dataset=train_dataset, batch_size=BATCH_SIZE,  
                               shuffle=True, num_workers=0)  
    test_loader = DataLoader(dataset=test_dataset, batch_size=BATCH_SIZE,  
                               shuffle=True, num_workers=0)      
    dataloaders = {}
    dataloaders['train'] = train_loader
    dataloaders['val'] = test_loader
    
    dataset_sizes = {}
    dataset_sizes['train'] = len(train_labels)
    dataset_sizes['val'] = len(test_labels)
    
    model_ft = models.resnet18(pretrained=True)
    num_ftrs = model_ft.fc.in_features
    model_ft.fc = nn.Linear(num_ftrs, 2)
    
    if use_gpu:
        model_ft = model_ft.cuda()
    
    criterion = nn.CrossEntropyLoss()
    
    # Observe that all parameters are being optimized
    optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)
    
    # Decay LR by a factor of 0.1 every 7 epochs
    exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)
    model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
                           num_epochs=25)
#    for (batch_x, batch_y) in enumerate(train_loader):
#        print(batch_x, batch_y)