-
Notifications
You must be signed in to change notification settings - Fork 0
/
googLenet_practice.py
222 lines (175 loc) · 7.65 KB
/
googLenet_practice.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
import numpy as np
import pandas as pd
import os
import torch
import torch.nn as nn
import matplotlib.pyplot as plt
import torchvision
from torch.utils.data import Dataset, DataLoader, ConcatDataset
from torchvision import transforms
from collections import OrderedDict
import torch.optim as optim
import torchvision.models as models
model = models.googlenet(pretrained = True)
device = "mps" if torch.backends.mps.is_available() else "cpu"
print(device)
for params in model.parameters():
#print(params)
params.requires_grad = False
#setting the model parameters to fix the data
model.fc = nn.Sequential(OrderedDict([
('fc1', nn.Linear(1024,2048)),
('relu', nn.ReLU()),
('fc2', nn.Linear(2048,2)),
('output', nn.LogSoftmax(dim = 1))
]))
#print(model)
#dataloader function
def load_data(data_folder, batch_size, num_workers):
transform = transforms.Compose([
transforms.Resize(256),
transforms.ColorJitter(),
transforms.RandomCrop(224),
transforms.RandomHorizontalFlip(),
transforms.Resize(224),
transforms.ToTensor()
])
data = torchvision.datasets.ImageFolder(root = data_folder, transform = transform)
data_loader = torch.utils.data.DataLoader(data, batch_size=batch_size, shuffle=True, num_workers = num_workers)
return data_loader
data_folder = '/Users/parkseongbeom/pytorch-test/deeplearning_practice/cat vs dog/train'
batch_size = 32
num_workers = 0
dataloader = load_data(data_folder, batch_size, num_workers)
"""#visualization
random_batch = random.choice(list(dataloader))
samples, labels = random_batch
# Generate random indices for images in the batch
num_images = 5 # Number of images to visualize
random_indices = random.sample(range(samples.shape[0]), num_images)
# Visualize the random images
fig, axes = plt.subplots(1, num_images, figsize=(15, 3))
for i, idx in enumerate(random_indices):
image = samples[idx].numpy().transpose((1, 2, 0))
label = labels[idx].item()
axes[i].imshow(image)
axes[i].set_title(f"Label: {label}")
axes[i].axis("off")
plt.tight_layout()
plt.show()"""
import torchvision.models as models
model = models.googlenet(pretrained = True)
# model part
model = model.to(device) #shifting model to gpu
loss = nn.CrossEntropyLoss()
loss1 = nn.CrossEntropyLoss()
loss2 = nn.CrossEntropyLoss()
discount = 0.3
optimizer = torch.optim.Adam(model.fc.parameters(), lr=0.001, amsgrad=True)
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[500,1000,1500], gamma=0.5)
epochs = 3
itr = 1
p_itr = 200
model.train()
total_loss = 0
loss_list = []
acc_list = []
for epoch in range(epochs):
for samples, labels in dataloader:
samples = samples.to(device)
samples = samples.to(torch.float32)
labels = labels.to(device)
#for param in model.parameters():
# param.requires_grad = True
#o,o1,o2 = model(samples) -> googlenet outputs using eager output
_loss = model(samples)
#check labels
loss_value = loss(_loss,labels)
#loss_value = loss(o,labels) + discount*(loss1(o1,labels) + loss2(o2,labels))
#loss = criterion(output[0], labels)
loss_value.backward()
optimizer.step()
total_loss += loss_value.item()
scheduler.step()
#output = torch.cat([o1, o2, o], dim=1)
output = total_loss
if itr%p_itr == 0:
pred = torch.argmax(output, dim=1)
correct = pred.eq(labels)
acc = torch.mean(correct.float())
print('Correct:{} pred:{} labels:[]'.format(correct,pred,labels))
print('[Epoch {}/{}] Iteration {} -> Train Loss: {:.4f}, Accuracy: {:.3f}'.format(epoch+1, epochs, itr, total_loss/p_itr, acc))
loss_list.append(total_loss/p_itr)
acc_list.append(acc)
total_loss = 0
itr += 1
# extract a feature with trained googlenet give it a condition to GAN
# train googlenet to extract features to identify cat and dogs
# with trained features extract inorder to use it as a condition for C-GAN
# to find out whether the models are trained give several inputs and test out whether the training holds and gan actually works
input_image = ''#!!
input_image = transforms(input_image)
input_image = input_image.unsqueeze(0) # Add batch dimension
num_epochs = 100
feature = model(input_image)
# Convert the extracted features to a tensor
conditional_data = feature.clone().detach()
import torch.nn as nn
# !!기초적인 GAN -> 확인해야함
class Generator(nn.Module):
def __init__(self, input_size, num_classes):
super(Generator, self).__init__()
self.fc = nn.Linear(input_size + num_classes, 1024)
# Add more layers as needed
def forward(self, features, conditional_data):
x = torch.cat([features, conditional_data], dim=1)
x = self.fc(x)
# Apply additional layers and transformations
return x
# Define the discriminator architecture
class Discriminator(nn.Module):
def __init__(self, input_size, num_classes):
super(Discriminator, self).__init__()
self.fc = nn.Linear(input_size + num_classes, 1)
# Add more layers as needed
def forward(self, image, features, conditional_data):
x = torch.cat([image, features, conditional_data], dim=1)
x = self.fc(x)
# Apply additional layers and transformations
return x
# !!check whether num_class holds
generator = Generator(input_size=model.fc.out_features, num_classes=2)
discriminator = Discriminator(input_size=model.fc.out_features, num_classes=2)
# !! 차이 확인, 해당 코드는 model feature를 직접적으로 c-gan의 condition으로 사용하려고 한다.
#generator = Generator(input_size=feature.size(1), num_classes=2)
#discriminator = Discriminator(input_size=feature.size(1), num_classes=2)
# Define loss function and optimizer
adversarial_loss = nn.BCEWithLogitsLoss()
generator_optimizer = optim.Adam(generator.parameters(), lr=0.0002, betas=(0.5, 0.999))
discriminator_optimizer = optim.Adam(discriminator.parameters(), lr=0.0002, betas=(0.5, 0.999))
# Training loop
for epoch in range(num_epochs):
for real_images, _ in dataloader: # Replace 'dataloader' with your actual data loading mechanism
batch_size = real_images.size(0)
# Generate fake images using the generator
fake_images = generator(feature, conditional_data)
# Train the discriminator
discriminator_real_outputs = discriminator(real_images, feature, conditional_data)
discriminator_fake_outputs = discriminator(fake_images.detach(), feature, conditional_data)
discriminator_real_loss = adversarial_loss(discriminator_real_outputs, torch.ones(batch_size, 1))
discriminator_fake_loss = adversarial_loss(discriminator_fake_outputs, torch.zeros(batch_size, 1))
discriminator_loss = discriminator_real_loss + discriminator_fake_loss
discriminator_optimizer.zero_grad()
discriminator_loss.backward()
discriminator_optimizer.step()
# Train the generator
discriminator_fake_outputs = discriminator(fake_images, feature, conditional_data)
generator_loss = adversarial_loss(discriminator_fake_outputs, torch.ones(batch_size, 1))
generator_optimizer.zero_grad()
generator_loss.backward()
generator_optimizer.step()
"""
With these changes, CGAN will now be conditioned on the features extracted from the trained GoogLeNet model.
The generator takes the features as input and generates fake images,
while the discriminator receives both real and fake images along with the corresponding features as input for discrimination.
"""