forked from luigifreda/pyslam
/
feature_contextdesc.py
223 lines (181 loc) · 9.43 KB
/
feature_contextdesc.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
"""
* This file is part of PYSLAM
*
* Adpated from https://github.com/lzx551402/contextdesc/blob/master/image_matching.py, see the license therein.
* Copyright (C) 2016-present Luigi Freda <luigi dot freda at gmail dot com>
*
* PYSLAM is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* PYSLAM is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with PYSLAM. If not, see <http://www.gnu.org/licenses/>.
"""
import config
config.cfg.set_lib('contextdesc',prepend=True)
from threading import RLock
import warnings # to disable tensorflow-numpy warnings: from https://github.com/tensorflow/tensorflow/issues/30427
warnings.filterwarnings('ignore', category=FutureWarning)
import os
import cv2
import numpy as np
if False:
import tensorflow as tf
else:
# from https://stackoverflow.com/questions/56820327/the-name-tf-session-is-deprecated-please-use-tf-compat-v1-session-instead
import tensorflow.compat.v1 as tf
from contextdesc.utils.opencvhelper import MatcherWrapper
#from contextdesc.models import get_model
from contextdesc.models.reg_model import RegModel
from contextdesc.models.loc_model import LocModel
from contextdesc.models.aug_model import AugModel
from utils_tf import set_tf_logging
#from utils_sys import Printer
kVerbose = True
# convert matrix of pts into list of keypoints
def convert_pts_to_keypoints(pts, scores, sizes):
assert(len(pts)==len(scores))
kps = []
if pts is not None:
# convert matrix [Nx2] of pts into list of keypoints
kps = [ cv2.KeyPoint(p[0], p[1], _size=sizes[i], _response=scores[i]) for i,p in enumerate(pts) ]
return kps
# interface for pySLAM
class ContextDescFeature2D:
quantize=False # Wheter to quantize or not the output descriptor
def __init__(self,
num_features=2000,
n_sample=2048, # Maximum number of sampled keypoints per octave
dense_desc=False, # Whether to use dense descriptor model
model_type='pb',
do_tf_logging=False):
print('Using ContextDescFeature2D')
self.lock = RLock()
self.model_base_path= config.cfg.root_folder + '/thirdparty/contextdesc/'
set_tf_logging(do_tf_logging)
self.num_features = num_features
self.n_sample = n_sample
self.model_type = model_type
self.dense_desc = dense_desc
self.quantize = ContextDescFeature2D.quantize
self.loc_model_path = self.model_base_path + 'pretrained/contextdesc++'
self.reg_model_path = self.model_base_path + 'pretrained/retrieval_model'
if self.model_type == 'pb':
reg_model_path = os.path.join(self.reg_model_path, 'reg.pb')
loc_model_path = os.path.join(self.loc_model_path, 'loc.pb')
aug_model_path = os.path.join(self.loc_model_path, 'aug.pb')
elif self.model_type == 'ckpt':
reg_model_path = os.path.join(self.reg_model_path, 'model.ckpt-550000')
loc_model_path = os.path.join(self.loc_model_path, 'model.ckpt-400000')
aug_model_path = os.path.join(self.loc_model_path, 'model.ckpt-400000')
else:
raise NotImplementedError
self.keypoint_size = 10 # just a representative size for visualization and in order to convert extracted points to cv2.KeyPoint
self.pts = []
self.kps = []
self.des = []
self.scales = []
self.scores = []
self.frame = None
print('==> Loading pre-trained network.')
self.ref_model = RegModel(reg_model_path) #get_model('reg_model')(reg_model_path) #RegModel(reg_model_path)
self.loc_model = LocModel(loc_model_path, **{'sift_desc': False, # compute or not SIFT descriptor (we do not need them here!)
'n_feature': self.num_features,
'n_sample': self.n_sample,
'peak_thld': 0.04,
'dense_desc': self.dense_desc,
'upright': False})
self.aug_model = AugModel(aug_model_path, **{'quantz': self.quantize})
print('==> Successfully loaded pre-trained network.')
def __del__(self):
with self.lock:
self.ref_model.close()
self.loc_model.close()
self.aug_model.close()
def prep_img(self,img):
rgb_list = []
gray_list = []
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)[..., np.newaxis]
img = img[..., ::-1]
rgb_list.append(img)
gray_list.append(gray)
return rgb_list, gray_list
# extract regional features
def extract_regional_features(self,rgb_list):
reg_feat_list = []
#model = get_model('reg_model')(model_path)
for _, val in enumerate(rgb_list):
reg_feat = self.ref_model.run_test_data(val)
reg_feat_list.append(reg_feat)
#model.close()
return reg_feat_list
# extract local features and keypoint matchability
def extract_local_features(self,gray_list):
cv_kpts_list = []
loc_info_list = []
loc_feat_list = []
sift_feat_list = []
# model = get_model('loc_model')(model_path, **{'sift_desc': True,
# 'n_sample': FLAGS.n_sample,
# 'peak_thld': 0.04,
# 'dense_desc': FLAGS.dense_desc,
# 'upright': False})
for _, val in enumerate(gray_list):
loc_feat, kpt_mb, normalized_xy, cv_kpts, sift_desc = self.loc_model.run_test_data(val)
raw_kpts = [np.array((i.pt[0], i.pt[1], i.size, i.angle, i.response)) for i in cv_kpts]
raw_kpts = np.stack(raw_kpts, axis=0)
loc_info = np.concatenate((raw_kpts, normalized_xy, loc_feat, kpt_mb), axis=-1)
cv_kpts_list.append(cv_kpts)
loc_info_list.append(loc_info)
sift_feat_list.append(sift_desc)
loc_feat_list.append(loc_feat / np.linalg.norm(loc_feat, axis=-1, keepdims=True))
#model.close()
return cv_kpts_list, loc_info_list, loc_feat_list, sift_feat_list
# extract augmented features
def extract_augmented_features(self,reg_feat_list, loc_info_list):
aug_feat_list = []
#model = get_model('aug_model')(model_path, **{'quantz': False})
assert len(reg_feat_list) == len(loc_info_list)
for idx, _ in enumerate(reg_feat_list):
aug_feat, _ = self.aug_model.run_test_data([reg_feat_list[idx], loc_info_list[idx]])
aug_feat_list.append(aug_feat)
#model.close()
return aug_feat_list
def compute_kps_des(self, frame):
with self.lock:
rgb_list, gray_list = self.prep_img(frame)
# extract regional features.
reg_feat_list = self.extract_regional_features(rgb_list)
# extract local features and keypoint matchability.
cv_kpts_list, loc_info_list, loc_feat_list, sift_feat_list = self.extract_local_features(gray_list)
# extract augmented features.
aug_feat_list = self.extract_augmented_features(reg_feat_list,loc_info_list)
self.kps = cv_kpts_list[0]
self.des = aug_feat_list[0]
return self.kps, self.des
def detectAndCompute(self, frame, mask=None): #mask is a fake input
with self.lock:
self.frame = frame
self.kps, self.des = self.compute_kps_des(frame)
if kVerbose:
print('detector: CONTEXTDESC, descriptor: CONTEXTDESC, #features: ', len(self.kps), ', frame res: ', frame.shape[0:2])
return self.kps, self.des
# return keypoints if available otherwise call detectAndCompute()
def detect(self, frame, mask=None): # mask is a fake input
with self.lock:
if self.frame is not frame:
self.detectAndCompute(frame)
return self.kps
# return descriptors if available otherwise call detectAndCompute()
def compute(self, frame, kps=None, mask=None): # kps is a fake input, mask is a fake input
with self.lock:
if self.frame is not frame:
#Printer.orange('WARNING: CONTEXTDESC is recomputing both kps and des on last input frame', frame.shape)
self.detectAndCompute(frame)
return self.kps, self.des