-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathdataset.py
144 lines (126 loc) · 7.39 KB
/
dataset.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
import os
import numpy as np
import torch
from torch.utils.data import DataLoader, Dataset, random_split
import torch.utils.data as data
import cv2
from models.utils.utils import get_BEV_tensor, get_BEV_projection
from models.utils.augment import train_transform
class VIGOR(Dataset):
def __init__(self, args, split='train', root = 'Data/VIGOR/', same_area=True):
usr = os.getcwd().split('/')[2]
root = os.path.join('/home',usr,root)
same_area = not args.cross_area
self.image_size = args.image_size
label_root = 'splits' # 'splits' splits__corrected
if same_area:
self.train_city_list =['NewYork', 'Seattle', 'SanFrancisco', 'Chicago'] # ['NewYork', 'Seattle', 'SanFrancisco', 'Chicago'] ['Seattle']
self.test_city_list = ['NewYork', 'Seattle', 'SanFrancisco', 'Chicago']
else:
self.train_city_list = ['NewYork', 'Seattle']
self.test_city_list = ['SanFrancisco', 'Chicago']
pano_list = []
pano_label = []
sat_delta = []
if split == 'train':
for city in self.train_city_list:
label_fname = os.path.join(root, label_root, city, 'same_area_balanced_train.txt'
if same_area else 'pano_label_balanced.txt')
with open(label_fname, 'r') as file:
for line in file.readlines():
data = np.array(line.split(' '))
label = []
for i in [1, 4, 7, 10]:
label.append(os.path.join(root, city, 'satellite', data[i]))
delta = np.array([data[2:4], data[5:7], data[8:10], data[11:13]]).astype(float)
pano_list.append(os.path.join(root, city, 'panorama', data[0]))
pano_label.append(label)
sat_delta.append(delta)
else:
for city in self.test_city_list:
label_fname = os.path.join(root, label_root, city, 'same_area_balanced_test.txt'
if same_area else 'pano_label_balanced.txt')
with open(label_fname, 'r') as file:
for line in file.readlines():
data = np.array(line.split(' '))
label = []
for i in [1, 4, 7, 10]:
label.append(os.path.join(root, city, 'satellite', data[i]))
delta = np.array([data[2:4], data[5:7], data[8:10], data[11:13]]).astype(float)
pano_list.append(os.path.join(root, city, 'panorama', data[0]))
pano_label.append(label)
sat_delta.append(delta)
self.pano_list = pano_list
self.pano_label = pano_label
self.sat_delta = sat_delta
self.split = split
self.transform = train_transform(0) if 'augment' in args and args.augment else None
self.center = [(self.image_size/2,self.image_size/2), (self.image_size/2,self.image_size/2-self.image_size/8)] \
if 'orien' in args and args.orien else [(self.image_size//2.0, self.image_size//2.0),]
pona_path = self.pano_list[0]
pona = cv2.imread(pona_path, 1)[:,:,::-1] # BGR ==> RGB
self.out = get_BEV_projection(pona,self.image_size,self.image_size,Fov = 85*2, dty = 0, dy = 0)
self.ori_noise = args.ori_noise
# self.out = None
def __len__(self):
return len(self.pano_list)
def __getitem__(self, idx):
patch_size = self.image_size
pona_path = self.pano_list[idx]
select_ = 0 #random.randint(0,3)
sat_path = self.pano_label[idx][select_]
pano_gps = np.array(pona_path[:-5].split(',')[-2:]).astype(float)
sat_gps = np.array(sat_path[:-4].split('_')[-2:]).astype(float)
# =================== read satellite map ===================================
sat = cv2.imread(sat_path, 1)[:,:,::-1]
sat = cv2.resize(sat, (patch_size, patch_size))
# =================== read ground map ===================================
pona = cv2.imread(pona_path, 1)[:,:,::-1]
rotation_range = self.ori_noise
random_ori = np.random.uniform(-1, 1) * rotation_range/360
ori_angle = random_ori * 360
pona = np.roll(pona,int(random_ori*pona.shape[1]), axis=1)
if self.split == 'train' and self.transform is not None:
pona_bev = get_BEV_tensor(pona,patch_size,patch_size,dty = 0, dy = 0, dataset=False, out = self.out).numpy().astype(np.uint8) # dataset=False get numpy HWC, else get tensor CHW
transformed = self.transform(image=pona_bev, keypoints=self.center)
pona_bev = transformed["image"]
try:
transformed_center = [transformed['keypoints'][0],transformed['keypoints'][1] ] \
if len(self.center) ==2 else transformed['keypoints']
except IndexError :
# if transformed_center is invisible, skip
print('\033[1;93m'+f"Skipping data at index {idx} due to invisible"+'. \033[0m')
return None
img1 = torch.from_numpy(pona_bev).float().permute(2, 0, 1)
else:
pona_bev = get_BEV_tensor(pona,500,500,dty = 0, dy = 0, out = self.out).numpy().astype(np.uint8)
pona_bev = cv2.resize(pona_bev, (patch_size, patch_size))
img1 = torch.from_numpy(pona_bev).float().permute(2, 0, 1)
img2 = torch.from_numpy(sat).float().permute(2, 0, 1)
pano_gps = torch.from_numpy(pano_gps) # [batch, 2]
sat_gps = torch.from_numpy(sat_gps)
sat_delta_init = torch.from_numpy(self.sat_delta[idx][select_]*patch_size/640.0).float()
sat_delta = torch.zeros(2)
sat_delta[1] = sat_delta_init[0] + patch_size/2.0
sat_delta[0] = patch_size/2.0 - sat_delta_init[1] # 从 [y, x] To [x, y], so fit the coord of model out
if self.split == 'train':
transformed_center = torch.tensor(transformed_center).float() if self.transform is not None else torch.tensor(self.center).float()
transformed_center = transformed_center.permute(1, 0)
return img1, img2, pano_gps, sat_gps, transformed_center, sat_delta, torch.tensor(ori_angle)
else:
return img1, img2, pano_gps, sat_gps, torch.tensor(ori_angle), sat_delta
def fetch_dataloader(args, split='train'):
train_dataset = VIGOR(args, split)
print('Training with %d image pairs' % len(train_dataset))
if split == 'train':
train_size = int(0.8 * len(train_dataset))
val_size = len(train_dataset) - train_size
train_dataset, val_dataset = random_split(train_dataset, [train_size, val_size])
print("using {} images for training, {} images for validation.".format(train_size, val_size))
return train_dataset, val_dataset
else:
nw = min([os.cpu_count(), args.batch_size if args.batch_size > 1 else 0, 8]) # number of workers
print('Using {} dataloader workers every process'.format(nw))
test_loader = data.DataLoader(train_dataset, batch_size=args.batch_size,
pin_memory=True, shuffle=True, num_workers=nw, drop_last=False)
return test_loader