-
Notifications
You must be signed in to change notification settings - Fork 4
/
data_loader.py
280 lines (217 loc) · 8.03 KB
/
data_loader.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
import chicken
import numpy as np
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms, datasets, models
from copy import deepcopy
import matplotlib.pyplot as plt
import random
def to_tensor(image):
trans = transforms.Compose([
transforms.ToTensor(),
])
return trans(image)
def show_tensor(tensor_image):
plt.imshow(tensor_image.permute(1, 2, 0))
plt.show()
def load_data(width_height=192, num_points=1, input_directory='data/inputs/', mask_directory='data/masks/'):
def simplify_masks(masks):
# Merge mask classes
BACKGROUND = np.array([0, 0, 170])
WALL = np.array([0, 0, 255])
DOOR = np.array([0, 170, 255])
WINDOW = np.array([0, 85, 255])
new_list = []
for picture in masks:
# new_picture = deepcopy(picture)
width, height, _ = picture.shape
new_picture = np.zeros((width, height, 4))
for x in range(width):
for y in range(height):
rgb = picture[x][y]
if np.allclose(rgb, WALL):
new_picture[x][y][1] = 1
elif np.allclose(rgb, DOOR):
new_picture[x][y][2] = 1
elif np.allclose(rgb, WINDOW):
new_rgb = WINDOW
new_picture[x][y][3] = 1
else:
new_picture[x][y][0] = 1
new_list.append(new_picture)
return new_list
print('Preparing input dataset...')
inputs = chicken.get_images(input_directory)
inputs = inputs + chicken.fliplr(inputs)
inputs = chicken.resize_and_smart_crop_square(inputs, width_height)
inputs = inputs[:num_points]
inputs = np.array(inputs)
print('Preparing mask dataset...')
masks = chicken.get_images(mask_directory)
masks = masks + chicken.fliplr(masks)
masks = chicken.resize_and_smart_crop_square(masks, width_height)
masks = masks[:num_points]
masks = simplify_masks(masks)
masks = np.array(masks)
print('Data loaded')
return inputs, masks
def get_PT_dataloaders(width_height=192, num_points=1, batch_size=1, input_directory='inputs_overfit/', label_directory='labels_overfit/'):
BACKGROUND = np.array([0, 0, 170])
WALL = np.array([0, 0, 255])
DOOR = np.array([0, 170, 255])
WINDOW = np.array([0, 85, 255])
classes = [
BACKGROUND, # BACKGROUND
WALL, # WALL
DOOR, # DOOR
WINDOW # WINDOW
]
classes = np.array(classes)
def simplify_label(label):
label = np.array(label)
width, height, _ = label.shape
one_hot_label = np.zeros((width, height, 4))
for x in range(width):
for y in range(height):
rgb = label[x][y]
if np.allclose(rgb, WALL):
one_hot_label[x][y][1] = 1
elif np.allclose(rgb, DOOR):
one_hot_label[x][y][2] = 1
elif np.allclose(rgb, WINDOW):
one_hot_label[x][y][3] = 1
else:
one_hot_label[x][y][0] = 1
return one_hot_label
preprocess_transform = transforms.Compose([
transforms.Resize(width_height),
transforms.CenterCrop(width_height),
transforms.ToTensor()
]) # TODO: Normalize with ImageNet mean/std
'''
T.Normalize(mean = [0.485, 0.456, 0.406],
std = [0.229, 0.224, 0.225])])
'''
lazy_label_transform = transforms.Compose([
transforms.ToPILImage(mode=None),
transforms.Lambda(simplify_label),
transforms.ToTensor()])
flip_transform = transforms.Compose([
transforms.ToPILImage(),
transforms.RandomHorizontalFlip(p=1.0),
transforms.ToTensor()])
class SegmentDataset(Dataset):
def __init__(self, inputs, masks, random_flip=True):
self.input_images = inputs
self.target_masks = masks
def __len__(self):
return len(self.input_images)
def __getitem__(self, index):
image = self.input_images[index]
mask = self.target_masks[index]
do_flip = random.random() > 0.5
# do_flip = True
if do_flip:
# image = transforms.ToPILImage()(image)
# image = transforms.RandomHorizontalFlip(p=1.0)(image)
# image = transforms.ToTensor()(image)
# mask = transforms.ToPILImage()(mask)
# mask = transforms.RandomHorizontalFlip(p=1.0)(mask)
# mask = transforms.ToTensor()(mask)
image = flip_transform(image)
mask = flip_transform(mask)
# image = transforms.functional.to_pil_image(image)
# image = transforms.functional.hflip(image)
# image = transforms.functional.to_tensor(image)
# mask = transforms.functional.to_pil_image(mask)
# mask = transforms.functional.hflip(mask)
# mask = transforms.functional.to_tensor(mask)
mask = lazy_label_transform(mask)
return [image, mask]
# Remove pytorch integer labels
print('Loading inputs...')
inputs = list(zip(*datasets.ImageFolder(root=input_directory, transform=preprocess_transform)))[0]
print('Loading labels...')
labels = list(zip(*datasets.ImageFolder(root=label_directory, transform=preprocess_transform)))[0]
# NUM_EVAL_IMAGES = len(inputs) // 10 # Designate 10% of dataset to val set
NUM_EVAL_IMAGES = max(len(inputs) // 10, 10) # Designate 10% of dataset to val set
if len(inputs) > NUM_EVAL_IMAGES:
train_images = inputs[:-NUM_EVAL_IMAGES]
train_labels = labels[:-NUM_EVAL_IMAGES]
val_images = inputs[-NUM_EVAL_IMAGES:]
val_labels = labels[-NUM_EVAL_IMAGES:]
else:
train_images = inputs
train_labels = labels
val_images = inputs
val_labels = labels
train_set = SegmentDataset(train_images, train_labels, random_flip=True)
# train_set = SegmentDataset(inputs[-NUM_EVAL_IMAGES:], labels[-NUM_EVAL_IMAGES:], random_flip=True)
val_set = SegmentDataset(val_images, val_labels, random_flip=True)
image_datasets = {
'train': train_set,
'val': val_set
}
dataloaders = {
'train': DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=0),
'val': DataLoader(val_set, batch_size=batch_size, shuffle=True, num_workers=0),
}
return dataloaders, image_datasets
def get_dataloaders(width_height=192, num_points=1, batch_size=1):
class SegmentDataset(Dataset):
def __init__(self, inputs, masks, transform=None):
self.input_images = np.array(inputs)
self.target_masks = np.array(masks)
self.transform = transform
def __len__(self):
return len(self.input_images)
def __getitem__(self, index):
image = self.input_images[index]
mask = self.target_masks[index]
if self.transform:
image = self.transform(image)
image = to_tensor(image)
mask = to_tensor(mask)
return [image, mask]
inputs, masks = load_data(width_height, num_points)
NUM_EVAL_IMAGES = max(len(inputs) // 10, 10) # Designate 10% of dataset to val set
if len(inputs) > NUM_EVAL_IMAGES:
train_images = inputs[:-NUM_EVAL_IMAGES]
train_labels = masks[:-NUM_EVAL_IMAGES]
val_images = inputs[-NUM_EVAL_IMAGES:]
val_labels = masks[-NUM_EVAL_IMAGES:]
else:
train_images = inputs
train_labels = masks
val_images = inputs
val_labels = masks
train_set = SegmentDataset(train_images, train_labels, transform=None)
val_set = SegmentDataset(val_images, val_labels, transform=None)
# train_set = SegmentDataset(inputs[:-NUM_EVAL_IMAGES], masks[:-NUM_EVAL_IMAGES], transform=None)
# # train_set = SegmentDataset(inputs[-NUM_EVAL_IMAGES:], masks[-NUM_EVAL_IMAGES:], transform=None)
# val_set = SegmentDataset(inputs[-NUM_EVAL_IMAGES:], masks[-NUM_EVAL_IMAGES:], transform=None)
image_datasets = {
'train': train_set,
'val': val_set
}
dataloaders = {
'train': DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=0),
'val': DataLoader(val_set, batch_size=batch_size, shuffle=True, num_workers=0),
}
print('Dataloaders created')
return dataloaders, image_datasets
def tensor_to_np(single_tensor):
# C, H, W
x = single_tensor.cpu().numpy()
x = x.transpose(1, 2, 0) # H, W, C
return x
def check_largest_dimensions(input_directory='data/masks/'):
inputs = chicken.get_images(input_directory)
largest_w = 0
largest_h = 0
for image in inputs:
shape = image.shape
width = shape[0]
height = shape[1]
largest_w = max(width, largest_w)
largest_h = max(height, largest_h)
print(largest_w, largest_h)