照搬pytorch官方代码，只是将数据集换成了INRIAPerson数据集中的train和test文件夹。
贴下代码和效果，代码是官方的，就不详细解释了。
# License: BSD
# Author: Sasank Chilamkurthy from __future__ import print_function, division 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 plt.ion() # interactive mode
# Data augmentation and normalization for training
# Just normalization for validation
data_transforms = {
'train': 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 = 'person'
image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x),
data_transforms[x])
for x in ['train', 'val']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=4,
shuffle=True, num_workers=4)
for x in ['train', 'val']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
class_names = image_datasets['train'].classes device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
def imshow(inp, title=None):
"""Imshow for Tensor."""
inp = inp.numpy().transpose((1, 2, 0))
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
inp = std * inp + mean
inp = np.clip(inp, 0, 1)
plt.imshow(inp)
if title is not None:
plt.title(title)
plt.pause(0.001) # pause a bit so that plots are updated # Get a batch of training data
inputs, classes = next(iter(dataloaders['train'])) # Make a grid from batch
out = torchvision.utils.make_grid(inputs) imshow(out, title=[class_names[x] for x in classes])
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 ['train', 'val']:
if phase == 'train':
scheduler.step()
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode running_loss = 0.0
running_corrects = 0 # 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 == 'train'):
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
loss = criterion(outputs, labels) # backward + optimize only if in training phase
if phase == 'train':
loss.backward()
optimizer.step() # statistics
running_loss += loss.item() * inputs.size(0)
running_corrects += torch.sum(preds == labels.data) 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)) # 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
def visualize_model(model, num_images=6):
was_training = model.training
model.eval()
images_so_far = 0
fig = plt.figure() with torch.no_grad():
for i, (inputs, labels) in enumerate(dataloaders['val']):
inputs = inputs.to(device)
labels = labels.to(device) outputs = model(inputs)
_, preds = torch.max(outputs, 1) for j in range(inputs.size()[0]):
images_so_far += 1
ax = plt.subplot(num_images//2, 2, images_so_far)
ax.axis('off')
ax.set_title('predicted: {}'.format(class_names[preds[j]]))
imshow(inputs.cpu().data[j]) if images_so_far == num_images:
model.train(mode=was_training)
return
model.train(mode=was_training)
model_ft = models.resnet18(pretrained=True)
num_ftrs = model_ft.fc.in_features
model_ft.fc = nn.Linear(num_ftrs, 2) 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)
model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
num_epochs=25)
Epoch 0/24
----------
train Loss: 0.4124 Acc: 0.8477
val Loss: 0.0737 Acc: 0.9744 Epoch 1/24
----------
train Loss: 0.2891 Acc: 0.9023
val Loss: 0.0836 Acc: 0.9703 Epoch 2/24
----------
train Loss: 0.3094 Acc: 0.9050
val Loss: 0.0614 Acc: 0.9771 Epoch 3/24
----------
train Loss: 0.2308 Acc: 0.9279
val Loss: 0.0429 Acc: 0.9865 Epoch 4/24
----------
train Loss: 0.1748 Acc: 0.9498
val Loss: 0.0331 Acc: 0.9906 Epoch 5/24
----------
train Loss: 0.2252 Acc: 0.9301
val Loss: 0.0702 Acc: 0.9906 Epoch 6/24
----------
train Loss: 0.1726 Acc: 0.9531
val Loss: 0.0442 Acc: 0.9852 Epoch 7/24
----------
train Loss: 0.1595 Acc: 0.9536
val Loss: 0.0359 Acc: 0.9906 Epoch 8/24
----------
train Loss: 0.1310 Acc: 0.9651
val Loss: 0.0355 Acc: 0.9892 Epoch 9/24
----------
train Loss: 0.1172 Acc: 0.9689
val Loss: 0.0325 Acc: 0.9906 Epoch 10/24
----------
train Loss: 0.1070 Acc: 0.9733
val Loss: 0.0515 Acc: 0.9838 Epoch 11/24
----------
train Loss: 0.1304 Acc: 0.9683
val Loss: 0.0452 Acc: 0.9892 Epoch 12/24
----------
train Loss: 0.1164 Acc: 0.9656
val Loss: 0.0424 Acc: 0.9892 Epoch 13/24
----------
train Loss: 0.0751 Acc: 0.9809
val Loss: 0.0396 Acc: 0.9906 Epoch 14/24
----------
train Loss: 0.1091 Acc: 0.9749
val Loss: 0.0279 Acc: 0.9946 Epoch 15/24
----------
train Loss: 0.0751 Acc: 0.9842
val Loss: 0.0352 Acc: 0.9906 Epoch 16/24
----------
train Loss: 0.1353 Acc: 0.9705
val Loss: 0.0413 Acc: 0.9879 Epoch 17/24
----------
train Loss: 0.0957 Acc: 0.9787
val Loss: 0.0332 Acc: 0.9906 Epoch 18/24
----------
train Loss: 0.1091 Acc: 0.9689
val Loss: 0.0317 Acc: 0.9906 Epoch 19/24
----------
train Loss: 0.1101 Acc: 0.9700
val Loss: 0.0402 Acc: 0.9879 Epoch 20/24
----------
train Loss: 0.1133 Acc: 0.9754
val Loss: 0.0392 Acc: 0.9892 Epoch 21/24
----------
train Loss: 0.0970 Acc: 0.9776
val Loss: 0.0424 Acc: 0.9865 Epoch 22/24
----------
train Loss: 0.0865 Acc: 0.9814
val Loss: 0.0348 Acc: 0.9919 Epoch 23/24
----------
train Loss: 0.1319 Acc: 0.9656
val Loss: 0.0341 Acc: 0.9892 Epoch 24/24
----------
train Loss: 0.0997 Acc: 0.9771
val Loss: 0.0328 Acc: 0.9906 Training complete in 9m 32s
Best val Acc: 0.994602
In [30]: visualize_model(model_ft)
visualize_model(model_ft)
model_conv = torchvision.models.resnet18(pretrained=True)
for param in model_conv.parameters():
param.requires_grad = False # Parameters of newly constructed modules have requires_grad=True by default
num_ftrs = model_conv.fc.in_features
model_conv.fc = nn.Linear(num_ftrs, 2) model_conv = model_conv.to(device) criterion = nn.CrossEntropyLoss() # Observe that only parameters of final layer are being optimized as
# opoosed to before.
optimizer_conv = optim.SGD(model_conv.fc.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_conv, step_size=7, gamma=0.1)
model_conv = train_model(model_conv, criterion, optimizer_conv,
exp_lr_scheduler, num_epochs=25)
Epoch 0/24
----------
train Loss: 0.3994 Acc: 0.8466
val Loss: 0.2137 Acc: 0.9109 Epoch 1/24
----------
train Loss: 0.2783 Acc: 0.8963
val Loss: 0.0649 Acc: 0.9744 Epoch 2/24
----------
train Loss: 0.2976 Acc: 0.8870
val Loss: 0.0577 Acc: 0.9811 Epoch 3/24
----------
train Loss: 0.2873 Acc: 0.9039
val Loss: 0.0477 Acc: 0.9825 Epoch 4/24
----------
train Loss: 0.3214 Acc: 0.8843
val Loss: 0.0499 Acc: 0.9798 Epoch 5/24
----------
train Loss: 0.3244 Acc: 0.8772
val Loss: 0.0483 Acc: 0.9798 Epoch 6/24
----------
train Loss: 0.2855 Acc: 0.8985
val Loss: 0.0446 Acc: 0.9825 Epoch 7/24
----------
train Loss: 0.2425 Acc: 0.9121
val Loss: 0.0460 Acc: 0.9798 Epoch 8/24
----------
train Loss: 0.2070 Acc: 0.9219
val Loss: 0.0390 Acc: 0.9879 Epoch 9/24
----------
train Loss: 0.2189 Acc: 0.9127
val Loss: 0.0408 Acc: 0.9825 Epoch 10/24
----------
train Loss: 0.2243 Acc: 0.9148
val Loss: 0.0577 Acc: 0.9825 Epoch 11/24
----------
train Loss: 0.2042 Acc: 0.9236
val Loss: 0.0519 Acc: 0.9852 Epoch 12/24
----------
train Loss: 0.2425 Acc: 0.9083
val Loss: 0.0440 Acc: 0.9838 Epoch 13/24
----------
train Loss: 0.2127 Acc: 0.9198
val Loss: 0.0454 Acc: 0.9865 Epoch 14/24
----------
train Loss: 0.2479 Acc: 0.9045
val Loss: 0.0551 Acc: 0.9771 Epoch 15/24
----------
train Loss: 0.2562 Acc: 0.8990
val Loss: 0.0491 Acc: 0.9852 Epoch 16/24
----------
train Loss: 0.2104 Acc: 0.9143
val Loss: 0.0448 Acc: 0.9852 Epoch 17/24
----------
train Loss: 0.2606 Acc: 0.8974
val Loss: 0.0480 Acc: 0.9798 Epoch 18/24
----------
train Loss: 0.2474 Acc: 0.9067
val Loss: 0.0639 Acc: 0.9798 Epoch 19/24
----------
train Loss: 0.2159 Acc: 0.9176
val Loss: 0.0495 Acc: 0.9852 Epoch 20/24
----------
train Loss: 0.2107 Acc: 0.9170
val Loss: 0.0482 Acc: 0.9838 Epoch 21/24
----------
train Loss: 0.2128 Acc: 0.9121
val Loss: 0.0522 Acc: 0.9838 Epoch 22/24
----------
train Loss: 0.2263 Acc: 0.9176
val Loss: 0.0459 Acc: 0.9852 Epoch 23/24
----------
train Loss: 0.1907 Acc: 0.9329
val Loss: 0.0460 Acc: 0.9906 Epoch 24/24
----------
train Loss: 0.2302 Acc: 0.9181
val Loss: 0.0425 Acc: 0.9879 Training complete in 4m 31s
Best val Acc: 0.990553
In [33]: visualize_model(model_conv)
visualize_model(model_conv) plt.ioff()
plt.show()
微调和特征提取两种方法的效果都很棒