-
Notifications
You must be signed in to change notification settings - Fork 0
/
classification.py
242 lines (198 loc) · 9.74 KB
/
classification.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
from matplotlib.ticker import MultipleLocator
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data.sampler import SubsetRandomSampler
import torchvision
import torchvision.transforms as transforms
import torchvision.models as models
from cls_dataset import CoinDataset
class Train(object):
def __init__(self, class_names, model, train_loader, valid_loader, test_loader, epochs, init_lr, valid_step=1):
self.train_loss_history = []
self.train_acc_history = []
self.valid_acc_history = []
self.classes_name = class_names
self.train_loader = train_loader
self.valid_loader = valid_loader
self.test_loader = test_loader
self.epochs = epochs
self.valid_step = valid_step
self.model = model
self.criterion = nn.CrossEntropyLoss()
self.optimizer = torch.optim.Adam(self.model.parameters(), lr=init_lr)
self.lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(self.optimizer, T_max=epochs-1, eta_min=5e-6)
self._run()
print('\n======================================================\n')
self._test()
def _calculate_acc(self, outputs, labels):
_, predicted = torch.max(outputs, 1)
acc = (predicted == labels).sum().item() / labels.shape[0]
return acc
def _valid_evaluation(self):
self.model.eval()
with torch.no_grad():
total_acc = 0
for data in self.valid_loader:
imgs, labels = data
imgs, labels = imgs.cuda(), labels.cuda()
outputs = self.model(imgs)
acc = self._calculate_acc(outputs.data, labels)
total_acc += acc
return total_acc / len(self.valid_loader)
def _test(self):
self.model.eval()
with torch.no_grad():
# total accuracy and class accuracy
class_correct = [0. for i in range(len(self.classes_name))]
class_total = [0. for i in range(len(self.classes_name))]
for data in self.test_loader:
imgs, labels = data
imgs, labels = imgs.cuda(), labels.cuda()
outputs = self.model(imgs)
_, predicted = torch.max(outputs.data, 1)
batch_correct_class = (predicted == labels)
for i in range(labels.shape[0]):
label = labels[i]
class_correct[label] += batch_correct_class[i].item()
class_total[label] += 1
for i in range(len(self.classes_name)):
print(f'Accuracy of {self.classes_name[i]}: {100 * class_correct[i] / class_total[i]} %')
print('')
print(f'Accuracy of the network on all testset: {100 * sum(class_correct) / sum(class_total)} %')
def plot_loss(self):
fig = plt.figure(figsize=(12,4))
ax = fig.gca()
plt.plot(list(range(1, self.epochs+1)), self.train_loss_history)
ax.xaxis.set_major_locator(MultipleLocator(1))
plt.title('Training loss')
plt.xlabel('epoch')
plt.ylabel('value')
plt.grid(color='r', linestyle='dotted', linewidth=1)
plt.xlim([1, self.epochs])
plt.show()
def plot_acc(self):
fig = plt.figure(figsize=(12,4))
ax = fig.gca()
acc_1, = plt.plot(list(range(1, self.epochs+1)), self.train_acc_history, '-b')
acc_2, = plt.plot(list(range(1, self.epochs+1)), self.valid_acc_history, '-g')
ax.xaxis.set_major_locator(MultipleLocator(1))
plt.title('Training and validation accuracy')
plt.xlabel('epoch')
plt.ylabel('value')
plt.legend([acc_1, acc_2], ['training accuracy', 'validation accuracy'], loc='lower right')
plt.grid(color='r', linestyle='dotted', linewidth=1)
plt.xlim([1, self.epochs])
plt.show()
def _run(self):
for epoch in range(self.epochs):
total_loss = 0
total_acc = 0
self.model.train()
for iteration, data in enumerate(self.train_loader):
imgs, labels = data
imgs, labels = imgs.cuda(), labels.cuda()
outputs = self.model(imgs)
loss = self.criterion(outputs, labels)
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
total_loss += loss.item()
acc = self._calculate_acc(outputs.data, labels)
total_acc += acc
if (iteration % 100) == 0:
print(f"Epoch[{epoch}][{iteration}/{len(self.train_loader)-1}]: loss: {loss.item()} acc: {acc*100} .")
total_loss /= len(self.train_loader)
self.train_loss_history.append(total_loss)
total_acc = (total_acc/len(self.train_loader))*100
self.train_acc_history.append(total_acc)
print(f'Epoch[{epoch}]: loss: {total_loss} acc: {total_acc} .')
self.lr_scheduler.step()
if epoch % self.valid_step == 0 or epoch == self.epochs-1:
valid_acc = self._valid_evaluation()
self.valid_acc_history.append(valid_acc*100)
def plot_acc(epochs, curve_1, curve_2, title):
fig = plt.figure(figsize=(12,4))
ax = fig.gca()
acc_1, = plt.plot(list(range(1, epochs+1)), curve_1, '-b')
acc_2, = plt.plot(list(range(1, epochs+1)), curve_2, '-g')
ax.xaxis.set_major_locator(MultipleLocator(1))
plt.title(title)
plt.xlabel('epoch')
plt.ylabel('value')
plt.legend([acc_1, acc_2], ['resnet50', 'mobilenetv2'], loc='lower right', framealpha=0.5)
plt.grid(color='r', linestyle='dotted', linewidth=1)
plt.xlim([1, epochs])
plt.show()
def main():
train_transform = transforms.Compose([
transforms.Resize((96, 96)),
transforms.RandomHorizontalFlip(p=0.5),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
transforms.RandomErasing(p=0.5)
])
valid_transform = transforms.Compose([
transforms.Resize((96, 96)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
test_transform = transforms.Compose([
transforms.Resize((96, 96)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
trainset = CoinDataset('/USER-DEFINED-PATH/coins/Train/', transform=train_transform)
print(f'train_num_imgs: {len(trainset)}')
validationset = CoinDataset('/USER-DEFINED-PATH/coins/Validation/', transform=valid_transform)
print(f'validation_num_imgs: {len(validationset)}')
testset = CoinDataset('/USER-DEFINED-PATH/coins/Test/', transform=test_transform)
print(f'test_num_imgs: {len(testset)}')
train_loader = torch.utils.data.DataLoader(trainset, batch_size=32, num_workers=4, pin_memory=True)
valid_loader = torch.utils.data.DataLoader(validationset, batch_size=32, num_workers=4, pin_memory=True)
test_loader = torch.utils.data.DataLoader(testset, batch_size=32, num_workers=4, pin_memory=True)
# training setting
epochs = 30
init_lr = 0.0003
valid_step = 1
# resnet 50
resnet50_model = models.resnet50(pretrained=False, num_classes=len(trainset.class_names))
resnet50_model_dict = resnet50_model.state_dict()
resnet50_pretrained_weight = torch.load('pre-trained_models/resnet50-19c8e357.pth')
resnet50_pretrained_weight = {k:v for k, v in resnet50_pretrained_weight.items() if (k in resnet50_model_dict and 'fc' not in k)}
resnet50_model.load_state_dict(resnet50_pretrained_weight, strict=False)
resnet50_model = resnet50_model.cuda()
resnet50_model_train = Train(trainset.class_names, resnet50_model, train_loader, valid_loader, test_loader, epochs, init_lr, valid_step=valid_step)
resnet50_model_train.plot_loss()
plt.savefig('resnet50_1.png')
resnet50_model_train.plot_acc()
plt.savefig('resnet50_2.png')
# mobilenet v2
mobilenetv2_model = models.mobilenet_v2(pretrained=False, num_classes=len(trainset.class_names))
mobilenetv2_model_dict = mobilenetv2_model.state_dict()
mobilenetv2_pretrained_weight = torch.load('pre-trained_models/mobilenet_v2-b0353104.pth')
mobilenetv2_pretrained_weight = {k:v for k, v in mobilenetv2_pretrained_weight.items() if (k in mobilenetv2_model_dict and 'classifier' not in k)}
mobilenetv2_model.load_state_dict(mobilenetv2_pretrained_weight, strict=False)
mobilenetv2_model = mobilenetv2_model.cuda()
mobilenetv2_model_train = Train(trainset.class_names, mobilenetv2_model, train_loader, valid_loader, test_loader, epochs, init_lr, valid_step=valid_step)
mobilenetv2_model_train.plot_loss()
plt.savefig('mobilenetv2_1.png')
mobilenetv2_model_train.plot_acc()
plt.savefig('mobilenetv2_2.png')
# save model
torch.save(resnet50_model.state_dict(), 'trained-models/resnet50.pth')
torch.save(mobilenetv2_model.state_dict(), 'trained-models/mobilenetv2.pth')
# comparison
plot_acc(epochs, resnet50_model_train.train_loss_history, mobilenetv2_model_train.train_loss_history, 'Training loss')
plt.savefig('training_loss.png')
plot_acc(epochs, resnet50_model_train.train_acc_history, mobilenetv2_model_train.train_acc_history, 'Training accuracy')
plt.savefig('training_acc.png')
plot_acc(epochs, resnet50_model_train.valid_acc_history, mobilenetv2_model_train.valid_acc_history, 'Validation accuracy')
plt.savefig('validation_acc.png')
if __name__ == '__main__':
main()