-
Notifications
You must be signed in to change notification settings - Fork 8
/
Prepare_Data_Results.py
146 lines (123 loc) · 6.68 KB
/
Prepare_Data_Results.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
# -*- coding: utf-8 -*-
"""
Created on Thu Dec 19 18:07:33 2019
Prepare data for visualization
@author: jpeeples
"""
## Python standard libraries
from __future__ import print_function
from __future__ import division
import numpy as np
from sklearn.model_selection import train_test_split
from torch.utils.data.sampler import SubsetRandomSampler
## PyTorch dependencies
import torch
from torchvision import transforms
## Local external libraries
from Datasets.DTD_loader import DTD_data
from Datasets.MINC_2500 import MINC_2500_data
from Datasets.GTOS_mobile_single_size import GTOS_mobile_single_data
def Prepare_DataLoaders(Results_parameters, split,input_size=224):
Dataset = Network_parameters['Dataset']
data_dir = Network_parameters['data_dir']
# Data augmentation and normalization for training
# Just normalization and resize for test
# Data transformations as described in:
# http://openaccess.thecvf.com/content_cvpr_2018/papers/Xue_Deep_Texture_Manifold_CVPR_2018_paper.pdf
if not(Network_parameters['rotation']):
data_transforms = {
'train': transforms.Compose([
transforms.Resize(Network_parameters['resize_size']),
transforms.RandomResizedCrop(input_size,scale=(.8,1.0)),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'val': transforms.Compose([
transforms.Resize(Network_parameters['center_size']),
transforms.CenterCrop(input_size),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}
else:
data_transforms = {
'train': transforms.Compose([
transforms.Resize(Network_parameters['resize_size']),
transforms.RandomResizedCrop(input_size,scale=(.8,1.0)),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
'val': transforms.Compose([
transforms.Resize(Network_parameters['center_size']),
transforms.CenterCrop(input_size),
transforms.RandomAffine(Network_parameters['degrees']),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}
# Create training and test datasets
if Dataset=='DTD':
train_dataset = DTD_data(data_dir, data='train',
numset = split + 1,
img_transform=data_transforms['val'])
validation_dataset = DTD_data(data_dir, data = 'val',
numset = split + 1,
img_transform=data_transforms['val'])
test_dataset = DTD_data(data_dir, data = 'test',
numset = split + 1,
img_transform=data_transforms['val'])
# #Combine training and test datasets
# train_dataset = torch.utils.data.ConcatDataset((train_dataset,validation_dataset))
elif Dataset == 'MINC_2500':
train_dataset = MINC_2500_data(data_dir, data='train',
numset = split + 1,
img_transform=data_transforms['val'])
validation_dataset = MINC_2500_data(data_dir, data='val',
numset = split + 1,
img_transform=data_transforms['val'])
test_dataset = MINC_2500_data(data_dir, data = 'test',
numset = split + 1,
img_transform=data_transforms['val'])
else:
# Create training and test datasets
dataset = GTOS_mobile_single_data(data_dir, train = True,
image_size=Network_parameters['resize_size'],
img_transform=data_transforms['val'])
X = np.ones(len(train_dataset))
train_indices = []
val_indices = []
kf = KFold(n_splits=splits,shuffle=True,
random_state=Network_parameters['random_state'])
for train_index, val_index in kf.split(X):
train_indices.append(train_index)
val_indices.append(val_index)
train_dataset = torch.utils.data.Subset(dataset, indices['train'][split])
validation_dataset = torch.utils.data.Subset(dataset, indices['val'][split])
test_dataset = GTOS_mobile_single_data(data_dir, train = False,
img_transform=data_transforms['val'])
image_datasets = {'train': train_dataset, 'val': validaton_dataset,
'test': test_dataset}
#If training dataset is larger than number of images for TSNE, subsample
if len(image_datasets['train']) > Results_parameters['Num_TSNE_images']:
indices = np.arange(len(image_datasets['train']))
y = image_datasets['train'].targets
#Use stratified split to balance training validation splits,
#set random state to be same for each encoding method
_,_,_,_,_,TSNE_indices = train_test_split(y,y,indices,
stratify=y,
test_size = Results_parameters['Num_TSNE_images'],
random_state=split+1)
# Creating PT data samplers and loaders:
TSNE_sampler = {'train': SubsetRandomSampler(TSNE_indices), 'val': None,'test': None}
else:
TSNE_sampler = {'train': None, 'val': None, 'test': None}
# Create training and test dataloaders
dataloaders_dict = {x: torch.utils.data.DataLoader(image_datasets[x],
batch_size=Results_parameters['batch_size'][x],
shuffle=False,
sampler = TSNE_sampler[x],
num_workers=Results_parameters['num_workers'],
pin_memory=Results_parameters['pin_memory']) for x in ['train', 'val','test']}
return dataloaders_dict