-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathimagenet_16h.py
301 lines (275 loc) · 10.8 KB
/
imagenet_16h.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
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
import torch
import numpy as np
import os
import random
import pandas as pd
import sys
import torch
import sys
import torch.nn as nn
sys.path.append("../")
sys.path.append("../networks")
from networks.cnn import DenseNet121_CE
import torchvision.transforms as transforms
from datasetsdefer.generic_dataset import GenericImageExpertDataset
from .basedataset import BaseDataset
# https://osf.io/2ntrf/
# https://www.pnas.org/doi/10.1073/pnas.2111547119
class ImageNet16h(BaseDataset):
def __init__(
self,
use_data_aug,
data_dir,
noise_version,
test_split=0.2,
val_split=0.1,
batch_size=1000,
get_embeddings=False,
transforms=None,
):
"""
Must go to https://osf.io/2ntrf/ , click on OSF Storage, download zip, unzip it, and write the path of the folder in data_dir
data_dir: where to save files for model
noise_version: noise version to use from 080,095, 110,125 (From imagenet16h paper)
use_data_aug: whether to use data augmentation (bool)
test_split: percentage of test data
val_split: percentage of data to be used for validation (from training set)
batch_size: batch size for training
transforms: data transforms
"""
self.data_dir = data_dir
self.use_data_aug = use_data_aug
self.test_split = test_split
self.val_split = val_split
self.batch_size = batch_size
self.n_dataset = 16
self.train_split = 1 - test_split - val_split
self.transforms = transforms
self.noise_version = noise_version
self.get_embeddings = get_embeddings
self.d = 1024
if self.noise_version not in ["080", "095", "110", "125"]:
raise ValueError(
'Noise version not supported, only pick from ["080","095","110","125"]'
)
self.generate_data()
def generate_data(self):
"""
generate data for training, validation and test sets
"""
# check if the folder data_dir has everything we need
if not os.path.exists(
self.data_dir
+ "/Behavioral Data/human_only_classification_6per_img_export.csv"
):
raise ValueError(
"cant find csv, Please download the data from https://osf.io/2ntrf/ , unzip it, and construct the path of the folder in data_dir"
)
if not os.path.exists(
self.data_dir + "/Noisy Images/phase_noise_" + self.noise_version
):
raise ValueError(
"cant find image folder, Please download the data from https://osf.io/2ntrf/ , unzip it, and construct the path of the folder in data_dir"
)
# load the csv file
data_behavioral = pd.read_csv(
self.data_dir
+ "/Behavioral Data/human_only_classification_6per_img_export.csv"
)
data_behavioral = data_behavioral[
data_behavioral["noise_level"] == int(self.noise_version)
]
data_behavioral = data_behavioral[
[
"participant_id",
"image_id",
"image_name",
"image_category",
"participant_classification",
"confidence",
]
]
# get unique categories
categories = data_behavioral["image_category"].unique()
# get mapping from category to index
self.category_to_idx = {categories[i]: i for i in range(len(categories))}
imagenames_categories = dict(
zip(data_behavioral["image_name"], data_behavioral["image_category"])
)
# for each image name, get all the participant classifications
image_name_to_participant_classifications = {}
for image_name in data_behavioral["image_name"].unique():
image_name_to_participant_classifications[image_name] = data_behavioral[
data_behavioral["image_name"] == image_name
]["participant_classification"].values
# sample a single classification from the participant classifications
image_name_to_single_participant_classification = {}
for image_name in image_name_to_participant_classifications:
image_name_to_single_participant_classification[
image_name
] = np.random.choice(image_name_to_participant_classifications[image_name])
image_names = os.listdir(
self.data_dir + "/Noisy Images/phase_noise_" + self.noise_version
)
image_names = [x for x in image_names if x.endswith(".png")]
# remove png extension
image_names = [x[:-4] for x in image_names]
image_paths = np.array(
[
"/data/ml2/shared/mozannar/improved_deferral/data/osfstorage-archive/Noisy Images/phase_noise_080/"
+ x
+ ".png"
for x in image_names
]
)
# get label for image names
image_names_labels = np.array(
[self.category_to_idx[imagenames_categories[x]] for x in image_names]
)
# get prediction for image names
image_names_human_predictions = np.array(
[
self.category_to_idx[image_name_to_single_participant_classification[x]]
for x in image_names
]
)
transform_train = transforms.Compose(
[
transforms.Resize(224),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
),
]
)
transform_test = transforms.Compose(
[
transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
),
]
)
transform_test_tensor = transforms.Compose(
[
transforms.ToTensor()
]
)
test_size = int(self.test_split * len(image_paths))
val_size = int(self.val_split * len(image_paths))
train_size = len(image_paths) - test_size - val_size
random_seed = random.randrange(10000)
train_x, val_x, test_x = torch.utils.data.random_split(
image_paths,
[train_size, val_size, test_size],
generator=torch.Generator().manual_seed(random_seed),
)
train_y, val_y, test_y = torch.utils.data.random_split(
image_names_labels,
[train_size, val_size, test_size],
generator=torch.Generator().manual_seed(random_seed),
)
train_h, val_h, test_h = torch.utils.data.random_split(
image_names_human_predictions,
[train_size, val_size, test_size],
generator=torch.Generator().manual_seed(random_seed),
)
data_train = GenericImageExpertDataset(
train_x.dataset[train_x.indices],
train_y.dataset[train_y.indices],
train_h.dataset[train_h.indices],
transform_train,
to_open=True,
)
data_val = GenericImageExpertDataset(
val_x.dataset[val_x.indices],
val_y.dataset[val_y.indices],
val_h.dataset[val_h.indices],
transform_test,
to_open=True,
)
data_test = GenericImageExpertDataset(
test_x.dataset[test_x.indices],
test_y.dataset[test_y.indices],
test_h.dataset[test_h.indices],
transform_test,
to_open=True,
)
self.data_train_loader = torch.utils.data.DataLoader(
data_train,
batch_size=self.batch_size,
shuffle=False,
num_workers=0,
pin_memory=True,
)
self.data_val_loader = torch.utils.data.DataLoader(
data_val,
batch_size=self.batch_size,
shuffle=False,
num_workers=0,
pin_memory=True,
)
self.data_test_loader = torch.utils.data.DataLoader(
data_test,
batch_size=self.batch_size,
shuffle=False,
num_workers=0,
pin_memory=True,
)
if self.get_embeddings:
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model_linear = DenseNet121_CE(16).to(device)
model_linear.densenet121.classifier = nn.Sequential(*list(model_linear.densenet121.classifier.children())[:-1])
# get embeddings of train-val-test data
def get_embeddings(model, data_loader):
model.eval()
with torch.no_grad():
embeddings = []
for i, (x, y, h) in enumerate(data_loader):
x = x.to(device)
y = y.to(device)
h = h.to(device)
x = model(x)
embeddings.append(x.cpu().numpy())
return np.concatenate(embeddings, axis=0)
train_embeddings = torch.FloatTensor(get_embeddings(model_linear, self.data_train_loader))
val_embeddings = torch.FloatTensor(get_embeddings(model_linear, self.data_val_loader))
test_embeddings = torch.FloatTensor(get_embeddings(model_linear, self.data_test_loader))
data_train = torch.utils.data.TensorDataset(
train_embeddings,
torch.from_numpy(train_y.dataset[train_y.indices]),
torch.from_numpy(train_h.dataset[train_h.indices]),
)
data_val = torch.utils.data.TensorDataset(
val_embeddings,
torch.from_numpy(val_y.dataset[val_y.indices]),
torch.from_numpy(val_h.dataset[val_h.indices]),
)
data_test = torch.utils.data.TensorDataset(
test_embeddings,
torch.from_numpy(test_y.dataset[test_y.indices]),
torch.from_numpy(test_h.dataset[test_h.indices]),
)
self.data_train_loader = torch.utils.data.DataLoader(
data_train,
batch_size=1000,
shuffle=False,
num_workers=0,
pin_memory=True,
)
self.data_val_loader = torch.utils.data.DataLoader(
data_val,
batch_size=1000,
shuffle=False,
num_workers=0,
pin_memory=True,
)
self.data_test_loader = torch.utils.data.DataLoader(
data_test,
batch_size=1000,
shuffle=False,
num_workers=0,
pin_memory=True,
)