-
Notifications
You must be signed in to change notification settings - Fork 22
/
Copy pathdatasets.py
256 lines (210 loc) · 7.39 KB
/
datasets.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
import os
import torch
from torch.utils.data import DataLoader, Dataset
from torchvision import datasets
import torchvision.transforms as transforms
from torchvision.transforms import functional as F
import torchvision
import glob
import PIL
import math
import numpy as np
import zipfile
import time
from scipy.io import loadmat
def read_pose(name,flip=False):
P = loadmat(name)['angle']
P_x = -(P[0,0] - 0.1) + math.pi/2
if not flip:
P_y = P[0,1] + math.pi/2
else:
P_y = -P[0,1] + math.pi/2
P = torch.tensor([P_x,P_y],dtype=torch.float32)
return P
def read_pose_npy(name,flip=False):
P = np.load(name)
P_x = P[0] + 0.14
if not flip:
P_y = P[1]
else:
P_y = -P[1] + math.pi
P = torch.tensor([P_x,P_y],dtype=torch.float32)
return P
def transform_matrix_to_camera_pos(c2w,flip=False):
"""
Get camera position with transform matrix
:param c2w: camera to world transform matrix
:return: camera position on spherical coord
"""
c2w[[0,1,2]] = c2w[[1,2,0]]
pos = c2w[:, -1].squeeze()
radius = float(np.linalg.norm(pos))
theta = float(np.arctan2(-pos[0], pos[2]))
phi = float(np.arctan(-pos[1] / np.linalg.norm(pos[::2])))
theta = theta + np.pi * 0.5
phi = phi + np.pi * 0.5
if flip:
theta = -theta + math.pi
P = torch.tensor([phi,theta],dtype=torch.float32)
return P
class CATS(Dataset):
def __init__(self, img_size, **kwargs):
super().__init__()
self.img_size = img_size
self.real_pose = False
if 'real_pose' in kwargs and kwargs['real_pose'] == True:
self.real_pose = True
for i in range(10):
try:
self.data = glob.glob(os.path.join('datasets/cats','*.png'))
assert len(self.data) > 0, "Can't find data; make sure you specify the path to your dataset"
if self.real_pose:
self.pose = [os.path.join('datasets/cats/poses',f.split('/')[-1].replace('.png','_pose.npy')) for f in self.data]
break
except:
print('failed to load dataset, try %02d times'%i)
time.sleep(0.5)
self.transform = transforms.Compose(
[transforms.Resize((img_size, img_size), interpolation=1), transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])
def __len__(self):
return len(self.data)
def __getitem__(self, index):
X = PIL.Image.open(self.data[index])
X = self.transform(X)
flip = (torch.rand(1) < 0.5)
if flip:
X = F.hflip(X)
if self.real_pose:
P = read_pose_npy(self.pose[index], flip=flip)
else:
P = 0
return X, P
class CARLA(Dataset):
def __init__(self, img_size, **kwargs):
super().__init__()
self.img_size = img_size
self.real_pose = False
if 'real_pose' in kwargs and kwargs['real_pose'] == True:
self.real_pose = True
for i in range(10):
try:
self.data = glob.glob(os.path.join('datasets/carla','*.png'))
assert len(self.data) > 0, "Can't find data; make sure you specify the path to your dataset"
if self.real_pose:
self.pose = [os.path.join('datasets/carla/poses',f.split('/')[-1].replace('.png','_extrinsics.npy')) for f in self.data]
break
except:
print('failed to load dataset, try %02d times'%i)
time.sleep(0.5)
self.transform = transforms.Compose(
[transforms.Resize((img_size, img_size), interpolation=1), transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])
def __len__(self):
return len(self.data)
def __getitem__(self, index):
X = PIL.Image.open(self.data[index])
X = self.transform(X)
flip = (torch.rand(1) < 0.5)
if flip:
X = F.hflip(X)
if self.real_pose:
P = transform_matrix_to_camera_pos(np.load(self.pose[index]), flip=flip)
else:
P = 0
return X, P
class FFHQ(Dataset):
def __init__(self, img_size, **kwargs):
super().__init__()
self.img_size = img_size
self.real_pose = False
if 'real_pose' in kwargs and kwargs['real_pose'] == True:
self.real_pose = True
for i in range(10):
try:
self.data = glob.glob(os.path.join('datasets/ffhq','*.png'))
assert len(self.data) > 0, "Can't find data; make sure you specify the path to your dataset"
if self.real_pose:
self.pose = [os.path.join('datasets/ffhq/poses',f.split('/')[-1].replace('png','mat')) for f in self.data]
break
except:
print('failed to load dataset, try %02d times'%i)
time.sleep(0.5)
self.transform = transforms.Compose(
[transforms.Resize((img_size, img_size), interpolation=1), transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])
def __len__(self):
return len(self.data)
def __getitem__(self, index):
X = PIL.Image.open(self.data[index])
X = self.transform(X)
flip = (torch.rand(1) < 0.5)
if flip:
X = F.hflip(X)
if self.real_pose:
P = read_pose(self.pose[index],flip=flip)
else:
P = 0
return X, P
def get_dataset(name, subsample=None, batch_size=1, **kwargs):
dataset = globals()[name](**kwargs)
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=batch_size,
shuffle=True,
drop_last=True,
pin_memory=False,
num_workers=8
)
return dataloader, 3
def get_dataset_(dataset, subsample=None, batch_size=1, **kwargs):
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=batch_size,
shuffle=True,
drop_last=True,
pin_memory=False,
num_workers=8
)
return dataloader, 3
def get_dataset_distributed(name, world_size, rank, batch_size, **kwargs):
dataset = globals()[name](**kwargs)
sampler = torch.utils.data.distributed.DistributedSampler(
dataset,
num_replicas=world_size,
rank=rank,
)
dataloader = torch.utils.data.DataLoader(
dataset,
sampler=sampler,
batch_size=batch_size,
shuffle=False,
drop_last=True,
pin_memory=False,
num_workers=16,
persistent_workers=True,
)
return dataloader, 3
def get_dataset_distributed_(_dataset, world_size, rank, batch_size, **kwargs):
sampler = torch.utils.data.distributed.DistributedSampler(
_dataset,
num_replicas=world_size,
rank=rank,
)
dataloader = torch.utils.data.DataLoader(
_dataset,
sampler=sampler,
batch_size=batch_size,
shuffle=False,
drop_last=True,
pin_memory=False,
num_workers=16,
persistent_workers=True,
)
return dataloader, 3
if __name__ == '__main__':
import imageio
from tqdm import tqdm
dataset = FFHQ(64, **{'real_pose': True})
dataset, _ = get_dataset_(dataset)
for i, (image, pose) in tqdm(enumerate(dataset)):
print(pose * 180 / np.pi)
imageio.imwrite('test.png', ((image.squeeze().permute(1, 2, 0)*0.5+0.5)*255).type(torch.uint8))
break