3

我正在尝试从标记的 Google Streeview 数据中对大量作物训练 resnet18 图像分类器。我正在跟随本教程。我有两个数据集，一个大约 20k 图像，一个大约 100k 图像。两个数据集都以相同的格式存储，并且都已上传到各自的 Google Cloud Storage 存储桶。gcsfuse然后，我使用--implicit-dirs标志将这两个存储桶安装在我的 VM 的主目录中。

然后我在我的 Google Compute Engine VM 上运行我的train.py文件，该 VM 是从Google Cloud Marketplace 上的Deep Learning VM 映像创建的。该虚拟机有一个 vCPU、一个 Nvidia Tesla K80 GPU、3.75gb 内存和一个 100gb 永久性磁盘。

当我运行训练脚本时，除了将 dataset_dir 变量指向gcsfuseVM 上正确的 -mounted 目录外，我没有进行任何更改。

当我train.py在 100k crop 目录上运行时，它运行得相对较快，单个 epoch 大约需要 30 分钟。我在它运行时跳入top，CPU 利用率很高，保持在 90% 左右。

然而，使用相同的虚拟机，当我train.py在 20K 作物目录上运行时，它运行得更慢，一个 epoch 需要 6-7 小时，尽管数据集的大小更小。在这种情况下，CPU 利用率永远不会超过 5%。

我无法弄清楚是什么导致了减速，因为除了数据集之外，两次运行之间没有什么不同（据我所知），它们的格式相同。我使用pytorch具有相同线程数的相同数据加载器。两个 GCS 存储桶位于同一区域，us-west1与我的 VM 实例位于同一区域。

似乎一个存储桶相对于另一个存储桶是 IO 受限的，但我不知道为什么。

任何想法表示赞赏！

我的train.py文件在下面。

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy
from collections import defaultdict



data_transforms = {
    'Test': transforms.Compose([
        transforms.RandomResizedCrop(224),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
    'Val': transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
}


data_dir = 'home/gweld/sliding_window_dataset/'
image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x),
                                          data_transforms[x])
                  for x in ['Test', 'Val']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=4,
                                             shuffle=True, num_workers=4)
              for x in ['Test', 'Val']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['Test', 'Val']}
class_names = image_datasets['Test'].classes

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




def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
    since = time.time()

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

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

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

            running_loss = 0.0
            running_corrects = 0

            class_corrects = defaultdict(int)
            class_totals   = defaultdict(int)

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

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                # track history if only in train
                with torch.set_grad_enabled(phase == 'Test'):
                    outputs = model(inputs)
                    _, preds = torch.max(outputs, 1)
                    loss = criterion(outputs, labels)

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

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

                for index, pred in enumerate(preds):
                    actual = labels.data[index]
                    class_name = class_names[actual]

                    if actual == pred: class_corrects[class_name] += 1
                    class_totals[class_name] += 1

            epoch_loss = running_loss / dataset_sizes[phase]
            epoch_acc = running_corrects.double() / dataset_sizes[phase]

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

            if phase == 'Val':
                print("Validation Class Accuracies")

                for class_name in class_totals:
                    class_acc = float(class_corrects[class_name])
                    class_acc = class_acc/class_totals[class_name]

                    print("{:20}{}%".format(class_name, 100*class_acc))
                print("\n")

            # 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_elapsed = time.time() - since
    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




model_ft = models.resnet18(pretrained=True)
num_ftrs = model_ft.fc.in_features
model_ft.fc = nn.Linear(num_ftrs, 5) # last arg here, # classes? -gw

model_ft = model_ft.to(device)

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)


# Train and evaluate
# ^^^^^^^^^^^^^^^^^^

print('Beginning Training on {} train and {} val images.'.format(dataset_sizes['Test'], dataset_sizes['Val']))


model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
                       num_epochs=25)




torch.save(model_ft.state_dict(), 'models/test_run_resnet18.pt')
4

0 回答 0