forked from luigifreda/pyslam
-
Notifications
You must be signed in to change notification settings - Fork 0
/
feature_manager_adaptors.py
329 lines (286 loc) · 16.4 KB
/
feature_manager_adaptors.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
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
"""
* This file is part of PYSLAM
*
* 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 math
from enum import Enum
import numpy as np
import cv2
from concurrent.futures import ThreadPoolExecutor, as_completed, wait
from utils_img import img_blocks, img_mask_blocks
from utils_features import sat_num_features
from utils_sys import Printer
from pyramid import Pyramid, PyramidType
kVerbose = True
kAdaptorNumRowDivs = 5 #2, 3
kAdaptorNumColDivs = 5 #2, 3
kNumLevelsInitSigma = 20
kBlockAdaptorUseParallelComputations = True
kBlockAdaptorMaxNumWorkers = 4
kPyramidAdaptorUseParallelComputations = True
kPyramidAdaptorMaxNumWorkers = 4
if not kVerbose:
def print(*args, **kwargs):
pass
# BlockAdaptor divides the image in row_divs x col_divs cells and extracts features in each of these cells
class BlockAdaptor(object):
def __init__(self,
detector,
descriptor = None,
row_divs = kAdaptorNumRowDivs,
col_divs = kAdaptorNumColDivs,
do_parallel = kBlockAdaptorUseParallelComputations):
self.detector = detector
self.descriptor = descriptor
self.row_divs = row_divs
self.col_divs = col_divs
self.do_parallel = do_parallel # do parallel computations
self.is_detector_equal_to_descriptor = (self.detector == self.descriptor)
def detect(self, frame, mask=None):
if self.row_divs == 1 and self.col_divs == 1:
return self.detector.detect(frame, mask)
else:
if kVerbose:
print('BlockAdaptor ', self.row_divs, 'x', self.col_divs)
block_generator = img_mask_blocks(frame, mask, self.row_divs, self.col_divs)
kps_all = [] # list are thread-safe
def detect_block(b_m_i_j):
b, m, i, j = b_m_i_j
if kVerbose and False:
print('BlockAdaptor in block (',i,',',j,')')
kps = self.detector.detect(b, mask=m)
#print('adaptor: detected #features: ', len(kps), ' in block (',i,',',j,')')
for kp in kps:
#print('kp.pt before: ', kp.pt)
kp.pt = (kp.pt[0] + j, kp.pt[1] + i)
#print('kp.pt after: ', kp.pt)
kps_all.extend(kps)
if not self.do_parallel:
# process the blocks sequentially
for b, m, i, j in block_generator:
detect_block((b,m,i,j))
else:
with ThreadPoolExecutor(max_workers = 4) as executor:
executor.map(detect_block, block_generator) # automatic join() at the end of the `width` block
return np.array(kps_all)
def detectAndCompute(self, frame, mask=None):
if self.row_divs == 1 and self.col_divs == 1:
return self.detector.detectAndCompute(frame, mask)
else:
if kVerbose:
print('BlockAdaptor ', self.row_divs, 'x', self.col_divs)
block_generator = img_mask_blocks(frame, mask, self.row_divs, self.col_divs)
kps_all = []
des_all = []
kps_des_map = {} # (i,j) -> (kps,des)
def detect_and_compute_block(b_m_i_j):
b, m, i, j = b_m_i_j
if kVerbose and False:
print('BlockAdaptor in block (',i,',',j,')')
if self.is_detector_equal_to_descriptor:
kps, des = self.detector.detectAndCompute(b, mask=m)
else:
kps = self.detector.detect(b, mask=m)
kps, des = self.descriptor.compute(b, kps)
#print('adaptor: detected #features: ', len(kps), ' in block (',i,',',j,')')
# transform the points
for kp in kps:
#print('kp.pt before: ', kp.pt)
kp.pt = (kp.pt[0] + j, kp.pt[1] + i)
#print('kp.pt after: ', kp.pt)
kps_des_map[(i,j)] = (kps,des)
if not self.do_parallel:
# process the blocks sequentially
for b, m, i, j in block_generator:
detect_and_compute_block((b, m, i, j))
else:
with ThreadPoolExecutor(max_workers = kBlockAdaptorMaxNumWorkers) as executor:
executor.map(detect_and_compute_block, block_generator) # automatic join() at the end of the `width` block
# now merge the computed results
for ij,(kps,des) in kps_des_map.items():
kps_all.extend(kps)
if des is not None and len(des)>0:
if len(des_all)>0:
des_all = np.vstack([des_all, des])
else:
des_all = des
return np.array(kps_all), np.array(des_all)
# PyramidAdaptor generate a pyramid of num_levels images and extracts features in each of these images
# TODO: check if a point on one level 'overlaps' with a point on other levels or add such option (DONE by FeatureManager.kdt_nms() )
class PyramidAdaptor(object):
def __init__(self,
detector,
descriptor=None,
num_features=2000,
num_levels=4,
scale_factor=1.2,
sigma0=1.0, # N.B.: SIFT use 1.6 for this value
first_level=0,
pyramid_type=PyramidType.RESIZE,
use_block_adaptor=False,
do_parallel = kPyramidAdaptorUseParallelComputations,
do_sat_features_per_level = False):
self.detector = detector
self.descriptor = descriptor
self.num_features = num_features
self.is_detector_equal_to_descriptor = (self.detector == self.descriptor)
self.num_levels = num_levels
self.scale_factor = scale_factor
self.inv_scale_factor = 1./scale_factor
self.sigma0 = sigma0
self.first_level = first_level
self.pyramid_type = pyramid_type
self.use_block_adaptor = use_block_adaptor
self.do_parallel = do_parallel # do parallel computations
self.do_sat_features_per_level = do_sat_features_per_level # saturate number of features for each level
self.pyramid = Pyramid(num_levels=num_levels,
scale_factor=scale_factor,
sigma0=sigma0,
first_level=first_level,
pyramid_type=pyramid_type)
self.initSigmaLevels()
self.block_adaptor = None
if self.use_block_adaptor:
self.block_adaptor = BlockAdaptor(self.detector, self.descriptor, row_divs = kAdaptorNumRowDivs, col_divs = kAdaptorNumColDivs, do_parallel=False)
def initSigmaLevels(self):
num_levels = max(kNumLevelsInitSigma, self.num_levels)
self.scale_factors = np.zeros(num_levels)
self.inv_scale_factors = np.zeros(num_levels)
self.scale_factors[0]=1.0
# compute desired number of features per level (by using the scale factor)
self.num_features_per_level = np.zeros(num_levels,dtype=np.int)
num_desired_features_per_level = self.num_features*(1 - self.inv_scale_factor)/(1 - math.pow(self.inv_scale_factor, self.num_levels))
sum_num_features = 0
for level in range(self.num_levels-1):
self.num_features_per_level[level] = int(round(num_desired_features_per_level))
sum_num_features += self.num_features_per_level[level];
num_desired_features_per_level *= self.inv_scale_factor
self.num_features_per_level[self.num_levels-1] = max(self.num_features - sum_num_features, 0)
#print('num_features_per_level:',self.num_features_per_level)
if self.first_level==-1:
self.scale_factors[0]=1.0/self.scale_factor
self.inv_scale_factors[0]=1.0/self.scale_factors[0]
for i in range(1,num_levels):
self.scale_factors[i]=self.scale_factors[i-1]*self.scale_factor
self.inv_scale_factors[i]=1.0/self.scale_factors[i]
#print('self.inv_scale_factors: ', self.inv_scale_factors)
# detect on 'unfiltered' pyramid images ('unfiltered' meanining depends on the selected pyramid type)
def detect(self, frame, mask=None):
if self.num_levels == 1:
return self.detector.detect(frame, mask)
else:
#TODO: manage mask
if kVerbose:
print('PyramidAdaptor #levels:', self.num_levels,'(from',self.first_level,'), scale_factor:', self.scale_factor,', sigma0:', self.sigma0,', type:', self.pyramid_type.name)
self.pyramid.compute(frame)
kps_all = [] # list are thread-safe
def detect_level(scale,pyr_cur,i):
kps = []
if self.block_adaptor is None:
kps = self.detector.detect(pyr_cur)
else:
kps = self.block_adaptor.detect(pyr_cur)
if kVerbose and False:
print("PyramidAdaptor - level", i, ", shape: ", pyr_cur.shape)
for kp in kps:
#print('kp.pt before: ', kp.pt)
kp.pt = (kp.pt[0]*scale, kp.pt[1]*scale)
kp.size = kp.size*scale
kp.octave = i
#print('kp: ', kp.pt, kp.octave)
if self.do_sat_features_per_level:
kps, _ = sat_num_features(kps, None, self.num_features_per_level[i]) # experimental
kps_all.extend(kps)
if not self.do_parallel:
#print('sequential computations')
# process the blocks sequentially
for i in range(0,self.num_levels):
scale = self.scale_factors[i]
pyr_cur = self.pyramid.imgs[i]
detect_level(scale,pyr_cur,i)
else:
#print('parallel computations')
futures = []
with ThreadPoolExecutor(max_workers = 4) as executor:
for i in range(0,self.num_levels):
scale = self.scale_factors[i]
pyr_cur = self.pyramid.imgs[i]
futures.append(executor.submit(detect_level, scale, pyr_cur, i))
wait(futures) # wait all the task are completed
return np.array(kps_all)
# detect on 'unfiltered' pyramid images ('unfiltered' meanining depends on the selected pyramid type)
# compute descriptors on 'filtered' pyramid images ('filtered' meanining depends on the selected pyramid type)
def detectAndCompute(self, frame, mask=None):
if self.num_levels == 1:
return self.detector.detectAndCompute(frame, mask)
else:
if kVerbose:
print('PyramidAdaptor [dc] #levels:', self.num_levels,'(from',self.first_level,'), scale_factor:', self.scale_factor,', sigma0:', self.sigma0,', type:', self.pyramid_type.name)
self.pyramid.compute(frame)
kps_all = []
des_all = []
kps_des_map = {} # i -> (kps,des)
def detect_and_compute_level(scale, pyr_cur, pyr_cur_filtered, N, i):
kps = []
if self.block_adaptor is None:
#kps, des = self.detector.detectAndCompute(pyr_cur)
if self.is_detector_equal_to_descriptor:
kps, des = self.detector.detectAndCompute(pyr_cur)
else:
kps = self.detector.detect(pyr_cur)
#print('description of filtered')
kps, des = self.descriptor.compute(pyr_cur_filtered, kps)
else:
kps, des = self.block_adaptor.detectAndCompute(pyr_cur)
if kVerbose and False:
print("PyramidAdaptor - level", i, ", shape: ", pyr_cur.shape)
for kp in kps:
#print('before: kp.pt:', kp.pt,', size:',kp.size,', octave:',kp.octave,', angle:',kp.angle)
kp.pt = (kp.pt[0]*scale, kp.pt[1]*scale)
kp.size = kp.size*scale
kp.octave = i
#print('after: kp.pt:', kp.pt,', size:',kp.size,', octave:',kp.octave,', angle:',kp.angle)
if self.do_sat_features_per_level:
kps, des = sat_num_features(kps, des, N) # experimental
kps_des_map[i] = (kps,des)
if not self.do_parallel:
#print('sequential computations')
# process the blocks sequentially
for i in range(0,self.num_levels):
scale = self.scale_factors[i]
pyr_cur = self.pyramid.imgs[i]
pyr_cur_filtered = self.pyramid.imgs_filtered[i]
detect_and_compute_level(scale, pyr_cur, pyr_cur_filtered, self.num_features_per_level[i], i)
else:
#print('parallel computations')
futures = []
with ThreadPoolExecutor(max_workers = 4) as executor:
for i in range(0,self.num_levels):
scale = self.scale_factors[i]
pyr_cur = self.pyramid.imgs[i]
pyr_cur_filtered = self.pyramid.imgs_filtered[i]
futures.append(executor.submit(detect_and_compute_level, scale, pyr_cur, pyr_cur_filtered, self.num_features_per_level[i], i))
wait(futures) # wait all the task are completed
# now merge the computed results
for i,(kps,des) in kps_des_map.items():
kps_all.extend(kps)
if des is not None and len(des)>0:
if len(des_all)>0:
des_all = np.vstack([des_all, des])
else:
des_all = des
return np.array(kps_all), np.array(des_all)