質問編集履歴
1
エラーを省略してしまっていた。
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -10,220 +10,228 @@
|
|
|
10
10
|
|
|
11
11
|
### 発生している問題・エラーメッセージ
|
|
12
12
|
|
|
13
|
+
Traceback (most recent call last):
|
|
14
|
+
|
|
15
|
+
File "Arashi/arashi.py", line 52, in <module>
|
|
16
|
+
|
|
17
|
+
datasets = torch.utils.data.TensorDataset(img_datas, labels)
|
|
18
|
+
|
|
19
|
+
File "/home/selen/.pyenv/versions/3.7.3/lib/python3.7/site-packages/torch/utils/data/dataset.py", line 36, in __init__
|
|
20
|
+
|
|
13
21
|
assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors)
|
|
14
22
|
|
|
23
|
+
File "/home/selen/.pyenv/versions/3.7.3/lib/python3.7/site-packages/torch/utils/data/dataset.py", line 36, in <genexpr>
|
|
24
|
+
|
|
25
|
+
assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors)
|
|
26
|
+
|
|
15
27
|
AttributeError: 'list' object has no attribute 'size'
|
|
16
28
|
|
|
29
|
+
### 該当のソースコード
|
|
30
|
+
|
|
31
|
+
import torch, torchvision
|
|
32
|
+
|
|
33
|
+
import torch.nn as nn
|
|
34
|
+
|
|
35
|
+
import torch.nn.functional as F
|
|
36
|
+
|
|
37
|
+
from torchvision import transforms
|
|
38
|
+
|
|
39
|
+
import pytorch_lightning as pl
|
|
40
|
+
|
|
41
|
+
from pytorch_lightning import Trainer
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
from PIL import Image
|
|
46
|
+
|
|
47
|
+
import glob
|
|
48
|
+
|
|
49
|
+
|
|
50
|
+
|
|
51
|
+
fold_path = '/home/selen/downloads/'
|
|
52
|
+
|
|
53
|
+
imgs = []
|
|
54
|
+
|
|
55
|
+
for imgs_path in glob.glob(fold_path + '*'):
|
|
56
|
+
|
|
57
|
+
imgs.append(glob.glob(imgs_path + '/*'))
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
from torchvision.models import resnet18
|
|
62
|
+
|
|
63
|
+
resnet = resnet18(pretrained=True)
|
|
64
|
+
|
|
65
|
+
|
|
66
|
+
|
|
67
|
+
transform = transforms.Compose([
|
|
68
|
+
|
|
69
|
+
transforms.Resize((224, 224)),
|
|
70
|
+
|
|
71
|
+
transforms.ToTensor(),
|
|
72
|
+
|
|
73
|
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
|
74
|
+
|
|
75
|
+
])
|
|
76
|
+
|
|
77
|
+
|
|
78
|
+
|
|
79
|
+
labels = []
|
|
80
|
+
|
|
81
|
+
img_datas = torch.tensor([])
|
|
82
|
+
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
for i,imgs_arr in enumerate(imgs):
|
|
86
|
+
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
for img_path in imgs_arr:
|
|
90
|
+
|
|
91
|
+
labels.append(i)
|
|
92
|
+
|
|
93
|
+
img = Image.open(img_path)
|
|
94
|
+
|
|
95
|
+
tensor_img = transform(img)
|
|
96
|
+
|
|
97
|
+
tensor_img = tensor_img.unsqueeze(0)
|
|
98
|
+
|
|
99
|
+
img_datas = torch.cat([img_datas, tensor_img],dim=0)
|
|
100
|
+
|
|
101
|
+
|
|
102
|
+
|
|
103
|
+
datasets = torch.utils.data.TensorDataset(img_datas, labels)
|
|
104
|
+
|
|
105
|
+
|
|
106
|
+
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
n_train = int(len(datasets) * 0.85)
|
|
110
|
+
|
|
111
|
+
n_val = len(datasets) - n_train
|
|
112
|
+
|
|
113
|
+
torch.manual_seed(0)
|
|
114
|
+
|
|
115
|
+
train,val = torch.utils.data.random_split(datasets,[n_train,n_val])
|
|
116
|
+
|
|
117
|
+
|
|
118
|
+
|
|
119
|
+
class TrainNet(pl.LightningModule):
|
|
120
|
+
|
|
121
|
+
@pl.data_loader
|
|
122
|
+
|
|
123
|
+
def train_dataloader(self):
|
|
124
|
+
|
|
125
|
+
return torch.utils.data.DataLoader(train, self.batch_size,shuffle=True)
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
def training_step(self, batch, batch_nb):
|
|
130
|
+
|
|
131
|
+
x, t = batch
|
|
132
|
+
|
|
133
|
+
y = self.forward(x)
|
|
134
|
+
|
|
135
|
+
loss = self.lossfun(y, t)
|
|
136
|
+
|
|
137
|
+
results = {'loss': loss}
|
|
138
|
+
|
|
139
|
+
return results
|
|
140
|
+
|
|
141
|
+
|
|
142
|
+
|
|
143
|
+
class ValidationNet(pl.LightningModule):
|
|
144
|
+
|
|
145
|
+
|
|
146
|
+
|
|
147
|
+
@pl.data_loader
|
|
148
|
+
|
|
149
|
+
def val_dataloader(self):
|
|
150
|
+
|
|
151
|
+
return torch.utils.data.DataLoader(val, self.batch_size)
|
|
152
|
+
|
|
153
|
+
|
|
154
|
+
|
|
155
|
+
def validation_step(self, batch, batch_nb):
|
|
156
|
+
|
|
157
|
+
x, t = batch
|
|
158
|
+
|
|
159
|
+
y = self.forward(x)
|
|
160
|
+
|
|
161
|
+
loss = self.lossfun(y, t)
|
|
162
|
+
|
|
163
|
+
y_label = torch.argmax(y, dim=1)
|
|
164
|
+
|
|
165
|
+
acc = torch.sum(t == y_label) * 1.0 / len(t)
|
|
166
|
+
|
|
167
|
+
results = {'val_loss': loss, 'val_acc': acc}
|
|
168
|
+
|
|
169
|
+
return results
|
|
170
|
+
|
|
171
|
+
|
|
172
|
+
|
|
173
|
+
def validation_end(self, outputs):
|
|
174
|
+
|
|
175
|
+
avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
|
|
176
|
+
|
|
177
|
+
avg_acc = torch.stack([x['val_acc'] for x in outputs]).mean()
|
|
178
|
+
|
|
179
|
+
results = {'val_loss': avg_loss, 'val_acc': avg_acc}
|
|
180
|
+
|
|
181
|
+
return results
|
|
182
|
+
|
|
183
|
+
|
|
184
|
+
|
|
185
|
+
class FineTuningNet(TrainNet, ValidationNet):
|
|
186
|
+
|
|
187
|
+
|
|
188
|
+
|
|
189
|
+
def __init__(self, batch_size=256):
|
|
190
|
+
|
|
191
|
+
super().__init__()
|
|
192
|
+
|
|
193
|
+
self.batch_size = batch_size
|
|
194
|
+
|
|
195
|
+
self.conv = resnet18(pretrained=True)
|
|
196
|
+
|
|
197
|
+
self.fc1 = nn.Linear(1000, 100)
|
|
198
|
+
|
|
199
|
+
self.fc2 = nn.Linear(100, 5)
|
|
200
|
+
|
|
201
|
+
for param in self.conv.parameters():
|
|
202
|
+
|
|
203
|
+
param.requires_grad = False
|
|
204
|
+
|
|
205
|
+
|
|
206
|
+
|
|
207
|
+
def lossfun(self, y, t):
|
|
208
|
+
|
|
209
|
+
return F.cross_entropy(y, t)
|
|
210
|
+
|
|
211
|
+
|
|
212
|
+
|
|
213
|
+
def configure_optimizers(self):
|
|
214
|
+
|
|
215
|
+
return torch.optim.SGD(self.parameters(), lr=0.01)
|
|
216
|
+
|
|
217
|
+
|
|
218
|
+
|
|
219
|
+
def forward(self, x):
|
|
220
|
+
|
|
221
|
+
x = self.conv(x)
|
|
222
|
+
|
|
223
|
+
x = self.fc1(x)
|
|
224
|
+
|
|
225
|
+
x = F.relu(x)
|
|
226
|
+
|
|
227
|
+
x = self.fc2(x)
|
|
228
|
+
|
|
229
|
+
return x
|
|
230
|
+
|
|
17
231
|
```
|
|
18
232
|
|
|
19
233
|
|
|
20
234
|
|
|
21
|
-
### 該当のソースコード
|
|
22
|
-
|
|
23
|
-
import torch, torchvision
|
|
24
|
-
|
|
25
|
-
import torch.nn as nn
|
|
26
|
-
|
|
27
|
-
import torch.nn.functional as F
|
|
28
|
-
|
|
29
|
-
from torchvision import transforms
|
|
30
|
-
|
|
31
|
-
import pytorch_lightning as pl
|
|
32
|
-
|
|
33
|
-
from pytorch_lightning import Trainer
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
from PIL import Image
|
|
38
|
-
|
|
39
|
-
import glob
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
fold_path = '/home/selen/downloads/'
|
|
44
|
-
|
|
45
|
-
imgs = []
|
|
46
|
-
|
|
47
|
-
for imgs_path in glob.glob(fold_path + '*'):
|
|
48
|
-
|
|
49
|
-
imgs.append(glob.glob(imgs_path + '/*'))
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
from torchvision.models import resnet18
|
|
54
|
-
|
|
55
|
-
resnet = resnet18(pretrained=True)
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
transform = transforms.Compose([
|
|
60
|
-
|
|
61
|
-
transforms.Resize((224, 224)),
|
|
62
|
-
|
|
63
|
-
transforms.ToTensor(),
|
|
64
|
-
|
|
65
|
-
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
|
|
66
|
-
|
|
67
|
-
])
|
|
68
|
-
|
|
69
|
-
|
|
70
|
-
|
|
71
|
-
labels = []
|
|
72
|
-
|
|
73
|
-
img_datas = torch.tensor([])
|
|
74
|
-
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
for i,imgs_arr in enumerate(imgs):
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
for img_path in imgs_arr:
|
|
82
|
-
|
|
83
|
-
labels.append(i)
|
|
84
|
-
|
|
85
|
-
img = Image.open(img_path)
|
|
86
|
-
|
|
87
|
-
tensor_img = transform(img)
|
|
88
|
-
|
|
89
|
-
tensor_img = tensor_img.unsqueeze(0)
|
|
90
|
-
|
|
91
|
-
img_datas = torch.cat([img_datas, tensor_img],dim=0)
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
datasets = torch.utils.data.TensorDataset(img_datas, labels)
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
|
|
99
|
-
|
|
100
|
-
|
|
101
|
-
n_train = int(len(datasets) * 0.85)
|
|
102
|
-
|
|
103
|
-
n_val = len(datasets) - n_train
|
|
104
|
-
|
|
105
|
-
torch.manual_seed(0)
|
|
106
|
-
|
|
107
|
-
train,val = torch.utils.data.random_split(datasets,[n_train,n_val])
|
|
108
|
-
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
class TrainNet(pl.LightningModule):
|
|
112
|
-
|
|
113
|
-
@pl.data_loader
|
|
114
|
-
|
|
115
|
-
def train_dataloader(self):
|
|
116
|
-
|
|
117
|
-
return torch.utils.data.DataLoader(train, self.batch_size,shuffle=True)
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
def training_step(self, batch, batch_nb):
|
|
122
|
-
|
|
123
|
-
x, t = batch
|
|
124
|
-
|
|
125
|
-
y = self.forward(x)
|
|
126
|
-
|
|
127
|
-
loss = self.lossfun(y, t)
|
|
128
|
-
|
|
129
|
-
results = {'loss': loss}
|
|
130
|
-
|
|
131
|
-
return results
|
|
132
|
-
|
|
133
|
-
|
|
134
|
-
|
|
135
|
-
class ValidationNet(pl.LightningModule):
|
|
136
|
-
|
|
137
|
-
|
|
138
|
-
|
|
139
|
-
@pl.data_loader
|
|
140
|
-
|
|
141
|
-
def val_dataloader(self):
|
|
142
|
-
|
|
143
|
-
return torch.utils.data.DataLoader(val, self.batch_size)
|
|
144
|
-
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
def validation_step(self, batch, batch_nb):
|
|
148
|
-
|
|
149
|
-
x, t = batch
|
|
150
|
-
|
|
151
|
-
y = self.forward(x)
|
|
152
|
-
|
|
153
|
-
loss = self.lossfun(y, t)
|
|
154
|
-
|
|
155
|
-
y_label = torch.argmax(y, dim=1)
|
|
156
|
-
|
|
157
|
-
acc = torch.sum(t == y_label) * 1.0 / len(t)
|
|
158
|
-
|
|
159
|
-
results = {'val_loss': loss, 'val_acc': acc}
|
|
160
|
-
|
|
161
|
-
return results
|
|
162
|
-
|
|
163
|
-
|
|
164
|
-
|
|
165
|
-
def validation_end(self, outputs):
|
|
166
|
-
|
|
167
|
-
avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
|
|
168
|
-
|
|
169
|
-
avg_acc = torch.stack([x['val_acc'] for x in outputs]).mean()
|
|
170
|
-
|
|
171
|
-
results = {'val_loss': avg_loss, 'val_acc': avg_acc}
|
|
172
|
-
|
|
173
|
-
return results
|
|
174
|
-
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
class FineTuningNet(TrainNet, ValidationNet):
|
|
178
|
-
|
|
179
|
-
|
|
180
|
-
|
|
181
|
-
def __init__(self, batch_size=256):
|
|
182
|
-
|
|
183
|
-
super().__init__()
|
|
184
|
-
|
|
185
|
-
self.batch_size = batch_size
|
|
186
|
-
|
|
187
|
-
self.conv = resnet18(pretrained=True)
|
|
188
|
-
|
|
189
|
-
self.fc1 = nn.Linear(1000, 100)
|
|
190
|
-
|
|
191
|
-
self.fc2 = nn.Linear(100, 5)
|
|
192
|
-
|
|
193
|
-
for param in self.conv.parameters():
|
|
194
|
-
|
|
195
|
-
param.requires_grad = False
|
|
196
|
-
|
|
197
|
-
|
|
198
|
-
|
|
199
|
-
def lossfun(self, y, t):
|
|
200
|
-
|
|
201
|
-
return F.cross_entropy(y, t)
|
|
202
|
-
|
|
203
|
-
|
|
204
|
-
|
|
205
|
-
def configure_optimizers(self):
|
|
206
|
-
|
|
207
|
-
return torch.optim.SGD(self.parameters(), lr=0.01)
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
def forward(self, x):
|
|
212
|
-
|
|
213
|
-
x = self.conv(x)
|
|
214
|
-
|
|
215
|
-
x = self.fc1(x)
|
|
216
|
-
|
|
217
|
-
x = F.relu(x)
|
|
218
|
-
|
|
219
|
-
x = self.fc2(x)
|
|
220
|
-
|
|
221
|
-
return x
|
|
222
|
-
|
|
223
|
-
```
|
|
224
|
-
|
|
225
|
-
|
|
226
|
-
|
|
227
235
|
|
|
228
236
|
|
|
229
237
|
### 試したこと
|