forked from SIAnalytics/dual-hrnet
-
Notifications
You must be signed in to change notification settings - Fork 4
/
xview2.py
208 lines (183 loc) · 9.31 KB
/
xview2.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
import os
import json
import cv2
import numpy as np
import random
from shapely import wkt
from shapely.geometry import Polygon
import torch
from torch.utils.data import Dataset, DataLoader
from utils import preprocess
import torchvision.transforms as transforms
import multiprocessing
multiprocessing.set_start_method('spawn', True)
class XView2Dataset(Dataset):
"""xView2
input: Post image
target: pixel-wise classes
"""
dmg_type = {'background': 0, 'no-damage': 1, 'minor-damage': 2, 'major-damage': 3, 'destroyed': 4,
'un-classified': 255}
diaster_type = {'earthquake': 0, 'fire': 1, 'tsunami': 2, 'volcano': 3, 'wind': 4, 'flooding': 5}
def __init__(self, root_dir, rgb_bgr='rgb', preprocessing=None, mode='train'):
assert mode in ('train', 'test')
self.mode = mode
self.root = root_dir
assert rgb_bgr in ('rgb', 'bgr')
self.rgb = bool(rgb_bgr == 'rgb')
self.preprocessing = preprocessing
self.dirs = {'train_imgs': os.path.join(self.root, 'train', 'images'),
'train_labs': os.path.join(self.root, 'train', 'labels'),
'tier3_imgs': os.path.join(self.root, 'tier3', 'images'),
'tier3_labs': os.path.join(self.root, 'tier3', 'labels'),
'test_imgs': os.path.join(self.root, 'test', 'images')}
train_imgs = [s for s in os.listdir(self.dirs['train_imgs'])]
tier3_imgs = [s for s in os.listdir(self.dirs['tier3_imgs'])]
train_labs = [s for s in os.listdir(self.dirs['train_labs'])]
tier3_labs = [s for s in os.listdir(self.dirs['tier3_labs'])]
test_imgs = [s for s in os.listdir(self.dirs['test_imgs'])]
self.sample_files = []
self.neg_sample_files = []
if self.mode == 'train':
self.add_samples_train(self.dirs['train_imgs'], self.dirs['train_labs'], train_imgs, train_labs)
self.add_samples_train(self.dirs['tier3_imgs'], self.dirs['tier3_labs'], tier3_imgs, tier3_labs)
else:
for pre in os.listdir(self.dirs['test_imgs']):
if pre[:9] != 'test_pre_':
continue
img_id = pre[9:][:-4]
post = 'test_post_' + pre[9:]
assert post in test_imgs
files = {'img_id': img_id,
'pre_img': os.path.join(self.dirs['test_imgs'], pre),
'post_img': os.path.join(self.dirs['test_imgs'], post)}
self.sample_files.append(files)
if mode == 'test':
self.data_transforms = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
def add_samples_train(self, img_dirs, lab_dirs, imgs, labs):
for pre in os.listdir(img_dirs):
if pre[-17:] != '_pre_disaster.png':
continue
chop = pre[:-4].split('_')
img_id = '_'.join(chop[:2])
post = img_id + '_post_disaster.png'
pre_json = img_id + '_pre_disaster.json'
post_json = img_id + '_post_disaster.json'
assert post in imgs
assert pre_json in labs
assert post_json in labs
assert img_id not in self.sample_files
files = {'img_id': img_id,
'pre_img': os.path.join(img_dirs, pre),
'post_img': os.path.join(img_dirs, post),
'pre_json': os.path.join(lab_dirs, pre_json),
'post_json': os.path.join(lab_dirs, post_json)}
self.sample_files.append(files)
def get_sample_info(self, idx):
files = self.sample_files[idx]
pre_img = cv2.imread(files['pre_img'])
post_img = cv2.imread(files['post_img'])
if self.rgb:
pre_img = cv2.cvtColor(pre_img, cv2.COLOR_BGR2RGB)
post_img = cv2.cvtColor(post_img, cv2.COLOR_BGR2RGB)
pre_json = json.loads(open(files['pre_json']).read())
post_json = json.loads(open(files['post_json']).read())
sample = {'pre_img': pre_img, 'post_img': post_img, 'image_id': files['img_id'],
'im_width': post_json['metadata']['width'],
'im_height': post_json['metadata']['height'],
'disaster': post_json['metadata']['disaster_type'],
'pre_meta': {m: pre_json['metadata'][m] for m in pre_json['metadata']},
'post_meta': {m: post_json['metadata'][m] for m in post_json['metadata']},
'pre_builds': dict(), 'post_builds': dict(), 'builds': dict()}
for b in pre_json['features']['xy']:
buid = b['properties']['uid']
sample['pre_builds'][buid] = {p: b['properties'][p] for p in b['properties']}
poly = Polygon(wkt.loads(b['wkt']))
sample['pre_builds'][buid]['poly'] = list(poly.exterior.coords)
for b in post_json['features']['xy']:
buid = b['properties']['uid']
sample['post_builds'][buid] = {p: b['properties'][p] for p in b['properties']}
poly = Polygon(wkt.loads(b['wkt']))
sample['post_builds'][buid]['poly'] = list(poly.exterior.coords)
sample['builds'][buid] = {'poly': list(poly.exterior.coords),
'subtype': b['properties']['subtype']}
# sample['mask_img'] = self.make_mask_img(**sample)
return sample
def __getitem__(self, idx):
files = self.sample_files[idx]
pre_img = cv2.imread(files['pre_img'])
post_img = cv2.imread(files['post_img'])
if self.rgb:
pre_img = cv2.cvtColor(pre_img, cv2.COLOR_BGR2RGB)
post_img = cv2.cvtColor(post_img, cv2.COLOR_BGR2RGB)
if self.mode == 'train':
sample = self.get_sample_with_mask(files, pre_img, post_img)
sample['image_id'] = files['img_id']
if self.preprocessing is not None:
transformed = preprocess(sample['pre_img'], sample['post_img'], sample['mask_img'],
flip=self.preprocessing['flip'],
scale=self.preprocessing['scale'],
crop=self.preprocessing['crop'])
sample['pre_img'] = transformed[0]
sample['post_img'] = transformed[1]
sample['mask_img'] = transformed[2]
else:
pre_img = self.data_transforms(pre_img)
post_img = self.data_transforms(post_img)
sample = {'pre_img': pre_img, 'post_img': post_img, 'image_id': files['img_id']}
return sample
@staticmethod
def _get_building_from_json(post_json):
buildings = dict()
for b in post_json['features']['xy']:
buid = b['properties']['uid']
poly = Polygon(wkt.loads(b['wkt']))
buildings[buid] = {'poly': list(poly.exterior.coords),
'subtype': b['properties']['subtype']}
return buildings
def get_sample_with_mask(self, files, pre_img, post_img):
post_json = json.loads(open(files['post_json']).read())
sample = {'pre_img': pre_img, 'post_img': post_img, 'image_id': files['img_id'],
'disaster': self.diaster_type[post_json['metadata']['disaster_type']]}
buildings = self._get_building_from_json(post_json)
sample['mask_img'] = self.make_mask_img(**buildings)
return sample
def make_mask_img(self, **kwargs):
width = 1024
height = 1024
builings = kwargs
mask_img = np.zeros([height, width], dtype=np.uint8)
for dmg in self.dmg_type:
polys_dmg = [np.array(builings[p]['poly']).round().astype(np.int32).reshape(-1, 1, 2)
for p in builings if builings[p]['subtype'] == dmg]
cv2.fillPoly(mask_img, polys_dmg, [self.dmg_type[dmg]])
return mask_img
def show_sample(self, **kwargs):
pass
def __len__(self):
return len(self.sample_files)
if __name__ == '__main__':
root_path = "/mnt/Dataset/xView2/v2"
dataset = XView2Dataset(root_path, rgb_bgr='rgb', preprocessing={'flip': True, 'scale': None, 'crop': (513, 513)})
dataset_test = XView2Dataset(root_path, rgb_bgr='rgb',
preprocessing={'flip': False, 'scale': (0.8, 2.0), 'crop': (1024, 1024)})
n_samples = len(dataset)
n_train = int(n_samples * 0.85)
n_test = n_samples - n_train
trainset, testset = torch.utils.data.random_split(dataset, [n_train, n_test])
dataloader = DataLoader(trainset, batch_size=5, shuffle=True, num_workers=4)
for i in range(n_test):
sample = testset[i]
original_idx = testset.indices[i]
info = dataset.get_sample_info(original_idx)
info2 = dataset_test.get_sample_info(original_idx)
sample2 = dataset_test[original_idx]
print(i, original_idx, sample['disaster'], sample['image_id'], sample['post_img'].shape)
print(i, original_idx, sample2['disaster'], sample2['image_id'], sample2['post_img'].shape)
print(i, original_idx, info['disaster'], info['image_id'])
print(i, original_idx, info2['disaster'], info2['image_id'])
for i, samples in enumerate(dataloader):
print(i, samples['disaster'], samples['image_id'], samples['post_img'].shape)