-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdatasource.py
189 lines (157 loc) · 6.9 KB
/
datasource.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
import cv2
import os
import torch
import torch.utils.data as data
import torchvision.transforms as T
from torch.utils.data import DataLoader,Dataset
import random
from PIL import Image
from torch.autograd import Variable
import PIL.ImageOps
import torch.nn as nn
from torch import optim
import numpy as np
import torch.nn.functional as F
import torchvision.datasets as dset
import torch.multiprocessing
from facenet_pytorch import MTCNN, InceptionResnetV1
download = True
sharing_strategy = "file_system"
torch.multiprocessing.set_sharing_strategy(sharing_strategy)
def set_worker_sharing_strategy(worker_id: int) -> None:
torch.multiprocessing.set_sharing_strategy(sharing_strategy)
class Config():
path = "./data/train/"
training_dir = "./data/train/"
testing_dir = "./data/train/"
validation_dir = "./data/train/"
batch_size = 32
train_number_epochs = 100
# class DataSource(object):
# def __init__(self):
# raise NotImplementedError()
# def partitioned_by_rows(self, num_workers, test_reserve=.3):
# raise NotImplementedError()
# def sample_single_non_iid(self, weight=None):
# raise NotImplementedError()
# # You may want to have IID or non-IID setting based on number of your peers
# # by default, this code brings all dataset
# class MedMNIST(DataSource):
# def __init__(self):
# self.data_flag = 'pathmnist'
# info = INFO[self.data_flag]
# self.n_channels = info['n_channels']
# self.n_classes = len(info['label'])
# self.task = info['task']
# DataClass = getattr(medmnist, info['python_class'])
# # preprocessing
# data_transform = transforms.Compose([
# transforms.ToTensor(),
# transforms.Normalize(mean=[.5], std=[.5])
# ])
# # load the data
# train_dataset = DataClass(split='train', transform=data_transform, download=download)
# test_dataset = DataClass(split='test', transform=data_transform, download=download)
# self.pil_dataset = DataClass(split='train', download=download)
# # encapsulate data into dataloader form
# self.train_loader = data.DataLoader(dataset=train_dataset, batch_size=BATCH_SIZE, shuffle=True)
# self.valid_loader = data.DataLoader(dataset=train_dataset, batch_size=2*BATCH_SIZE, shuffle=False)
# self.test_loader = data.DataLoader(dataset=test_dataset, batch_size=2*BATCH_SIZE, shuffle=False)
# print(train_dataset)
# print("===================")
# print(test_dataset)
class DataSetFactory:
def __init__(self, type=None):
self.split_type = 'iid' if type == None else type
img = np.load(Config.path + "olivetti_faces.npy")
transforms = T.Compose([T.ToTensor()])
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
resnet = InceptionResnetV1(pretrained="vggface2").to(device).eval()
train_data = []
test_data = []
validate_data = []
# for i in range(25):
# train_data.append(img[i * 10 : i * 10 + 10])
train_data = self.split_train_data(self.split_type, img, 25)
for i in range(25,30):
validate_data.append(img[i * 10 : i * 10 + 10])
for i in range(30,40):
test_data.append(img[i * 10 : i * 10 + 10])
# make it 500 100 200
self.num_train = 200
self.num_valid = 10
self.num_test = 30
print('training size %d :validate size %d : test size %d' % (
self.num_train, self.num_valid, self.num_test))
training = olivetti_faces_dataset(images=train_data, transforms=transforms, size=self.num_train, resnet=resnet, device=device)
test = olivetti_faces_dataset(images=test_data, transforms=transforms, size=self.num_test, resnet=resnet, device=device)
validate = olivetti_faces_dataset(images=validate_data, transforms=transforms, size=self.num_valid, resnet=resnet, device=device)
self.train_loader = DataLoader(training, batch_size=Config.batch_size, shuffle=True, worker_init_fn=set_worker_sharing_strategy)
self.valid_loader = DataLoader(validate, batch_size=Config.batch_size, shuffle=True, worker_init_fn=set_worker_sharing_strategy)
self.test_loader = DataLoader(test, batch_size=Config.batch_size, shuffle=True, worker_init_fn=set_worker_sharing_strategy)
def split_train_data(self, type, img, max_range):
data = []
if(type == 'iid'):
first = random.randrange(0, max_range)
second = first
third = first
while(second == first):
second = random.randrange(0, max_range)
while(third == first):
third = random.randrange(0, max_range)
data.append(img[first * 10 : first * 10 + 10])
data.append(img[second * 10 : second * 10 + 10])
data.append(img[third * 10 : third * 10 + 10])
else:
first = random.randrange(0, max_range)
second = first
while(second == first):
second = random.randrange(0, max_range)
data.append(img[first * 10 : first * 10 + 10])
data[0][0] = img[second][2]
data[0][1] = img[second][7]
random.shuffle(data[0])
return data
class olivetti_faces_dataset(Dataset):
def __init__(self, images, transforms, size, resnet, device):
self.data = images
self.transforms = transforms
self.size = size
self.resnet = resnet
self.device = device
def __len__(self):
return self.size
def embedding(self, x):
with torch.no_grad():
x = cv2.merge((x, x, x)) * 255
x = cv2.resize(x, dsize=(128, 128), interpolation=cv2.INTER_CUBIC)
x = (x - 127.5) / 128.0
return self.resnet(self.transforms(x).unsqueeze(0).to(self.device))
def __getitem__(self, idx):
img1, img2, label = None, None, None
if idx % 2 == 0: # same img
label = 1
n = random.randrange(0, len(self.data))
img1 = self.data[n][random.randrange(0, 10)]
img2 = self.data[n][random.randrange(0, 10)]
else:
label = 0
n = random.randrange(0, len(self.data))
img1 = self.data[n][random.randrange(0, 10)]
n = random.randrange(0, len(self.data))
img2 = self.data[n][random.randrange(0, 10)]
img1_embedding = self.embedding(img1)
img2_embedding = self.embedding(img2)
return (
img1_embedding,
img2_embedding,
torch.FloatTensor([label]).to(self.device),
)
if __name__ == "__main__":
m = DataSetFactory()
# for i,data in enumerate(m.train_loader,0):
# print (data[2].size(0))
# # print(i)
# # img0, img1 , label = data
# # print(type(img0),type(img1), type(label))
# # print(label)