forked from chrischoy/open-ucn
-
Notifications
You must be signed in to change notification settings - Fork 0
/
extract_feature.py
179 lines (149 loc) · 5.3 KB
/
extract_feature.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
import argparse
import glob
import itertools
import os.path as osp
import time
from multiprocessing import Pool as ThreadPool
from urllib.request import urlretrieve
import cv2
import h5py
import numpy as np
import torch
from lib.timer import Timer
from lib.util_2d import normalize_keypoint
from model.resunet import ResUNetBN2D2
from util.file import loadh5
from util.visualization import visualize_image_correspondence
def get_pool_result(num_processor, fun, args):
pool = ThreadPool(num_processor)
pool_res = pool.map(fun, args)
pool.close()
pool.join()
return pool_res
def prep_image(full_path):
assert osp.exists(full_path), f"File {full_path} does not exist."
return cv2.imread(full_path, cv2.IMREAD_GRAYSCALE)
def to_normalized_torch(img, device):
"""
Normalize the image to [-0.5, 0.5] range and augment batch and channel dimensions.
"""
img = img.astype(np.float32) / 255 - 0.5
return torch.from_numpy(img).to(device)[None, None, :, :]
def random_sample(arr, n):
np.random.seed(0)
total = len(arr)
num_sample = min(total, n)
idx = sorted(np.random.choice(range(total), num_sample, replace=False))
return np.asarray(arr)[idx]
def dump_correspondence_single(args):
img0, img1, calib_path0, calib_path1, F0, F1, i, len_dset, source = args
calib0 = loadh5(osp.join(source, calib_path0))
calib1 = loadh5(osp.join(source, calib_path1))
K0, K1 = calib0['K'], calib1["K"]
imsize0, imsize1 = calib0['imsize'], calib1['imsize']
# build correspondences
x0, y0, x1, y1 = visualize_image_correspondence(
img0, img1, F0[0], F1[0], i, mode='gpu-all', config=config, visualize=False)
kp0 = np.stack((x0, y0), 1).astype(np.float32)
kp1 = np.stack((x1, y1), 1).astype(np.float32)
norm_kp0 = normalize_keypoint(kp0, K0, imsize0 * 0.5)[:, :2]
norm_kp1 = normalize_keypoint(kp1, K1, imsize1 * 0.5)[:, :2]
coords = np.concatenate((kp0, kp1), axis=1)
n_coords = np.concatenate((norm_kp0, norm_kp1), axis=1)
return coords, n_coords
def dump_correspondences(config):
# load model
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
checkpoint = torch.load(config.weights)
model = ResUNetBN2D2(1, 64, normalize_feature=True)
model.load_state_dict(checkpoint['state_dict'])
model.eval()
model = model.to(device)
print("load model")
# load dataset
source = config.source
with h5py.File(config.target, 'r+') as ofp:
new_data = {}
keys = ['ucn_coords', 'ucn_n_coords']
for k in keys:
if k in ofp.keys():
new_data[k] = ofp[k]
else:
new_data[k] = ofp.create_group(k)
len_dset = len(ofp['coords'])
keys = ofp['ucn_coords'].keys()
print("len dataset : ", len_dset)
matching_timer, write_timer = Timer(), Timer()
# extract correspondences
for i in range(len_dset):
# skip existing pair
if str(i) in keys:
continue
_coords = ofp['coords'][str(i)]
img_path0 = _coords.attrs['img0']
img_path1 = _coords.attrs['img1']
img_idx0 = int(_coords.attrs['idx0']) + 1
img_idx1 = int(_coords.attrs['idx1']) + 1
calib_path0 = "/".join(img_path0.split("/")[:-2])
calib_path0 += f"/calibration/calibration_{img_idx0:06d}.h5"
calib_path1 = "/".join(img_path1.split("/")[:-2])
calib_path1 += f"/calibration/calibration_{img_idx1:06d}.h5"
img0 = prep_image(osp.join(source, img_path0))
img1 = prep_image(osp.join(source, img_path1))
F0 = model(to_normalized_torch(img0, device))
F1 = model(to_normalized_torch(img1, device))
args = (img0, img1, calib_path0, calib_path1, F0, F1, i, len_dset, source)
matching_timer.tic()
coords, n_coords = dump_correspondence_single(args)
matching_timer.toc()
write_timer.tic()
coords_data = new_data['ucn_coords'].create_dataset(
str(i), coords.shape, dtype=np.float32)
coords_data[:] = coords.astype(np.float32)
n_coords_data = new_data['ucn_n_coords'].create_dataset(
str(i), n_coords.shape, dtype=np.float32)
n_coords_data[:] = n_coords.astype(np.float32)
write_timer.toc()
print(
f"[{i}/{len_dset}] save {coords.shape} coordinate, matching {matching_timer.avg:.3f}, write {write_timer.avg:.3f}"
)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument(
'--weights',
default='ResUNetBN2D2-YFCC100train.pth',
type=str,
help='Path to pretrained weights')
parser.add_argument(
'--source',
type=str,
required=True,
help="source directory of YFCC100M dataset",
)
parser.add_argument(
'--target',
type=str,
required=True,
help="target directory to save processed data",
)
parser.add_argument(
'--num_kp',
type=int,
default=10000,
)
parser.add_argument(
'--ucn_inlier_threshold_pixel',
type=float,
default=4,
help="Inlier threshold for hit test")
parser.add_argument('--num_processor', type=int, default=8)
config = parser.parse_args()
print(config)
if not osp.isfile('ResUNetBN2D2-YFCC100train.pth'):
print('Downloading weights...')
urlretrieve(
"https://node1.chrischoy.org/data/publications/ucn/ResUNetBN2D2-YFCC100train-100epoch.pth",
'ResUNetBN2D2-YFCC100train.pth')
print("start")
with torch.no_grad():
dump_correspondences(config)