-
Notifications
You must be signed in to change notification settings - Fork 3
/
extract_keypoints.py
executable file
·441 lines (324 loc) · 14.7 KB
/
extract_keypoints.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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
"""
PROGRAMA PARA CALCULAR KEYPOINTS DE DOS IMAGENES 512x1024, ASI MISMO DE SUS CORRESPONDENCIAS
POR UN KNN BILATERAL
"""
import time
import torch
import torch.nn.functional as F
from spherical_distortion.functional import create_tangent_images, unresample
from spherical_distortion.util import *
import matplotlib.pyplot as plt
from skimage import io
import os
import cv2
import sys
import pandas as pd
import numpy as np
import _spherical_distortion_ext._mesh as _mesh
import argparse
from random import sample
import imageio
from scipy.spatial.transform import Rotation as Rot
from utils.coord import coord_3d
from utils.ransac import *
from utils.keypoint import *
from utils.metrics import *
from utils.camera_recovering import *
from os import listdir
from os.path import isfile, join, isdir
from tqdm import tqdm
sys.path.append(os.getcwd()+'/SPHORB-master')
import build1.sphorb_cpp as sphorb
def sort_key(pts1, pts2, desc1, desc2, points):
ind1 = np.argsort(pts1[:,2].numpy(),axis = 0)[::-1]
ind2 = np.argsort(pts2[:,2].numpy(),axis = 0)[::-1]
max1 = np.min([points,ind1.shape[0]])
max2 = np.min([points,ind2.shape[0]])
ind1 = ind1[:max1]
ind2 = ind2[:max2]
pts1 = pts1[ind1.copy(),:]
pts2 = pts2[ind2.copy(),:]
desc1 = desc1[:,ind1.copy()]
desc2 = desc2[:,ind2.copy()]
pts1 = np.concatenate((pts1[:,:2], np.ones((pts1.shape[0],1))), axis = 1 )
pts2 = np.concatenate((pts2[:,:2], np.ones((pts2.shape[0],1))), axis = 1 )
desc1 = np.transpose(desc1,[1,0]).numpy()
desc2 = np.transpose(desc2,[1,0]).numpy()
return pts1, pts2, desc1, desc2
def matched_points(pts1, pts2, desc1, desc2, opt, args_opt, match='ratio'):
if opt[-1] == 'p':
porce = int(opt[:-1])
n_key = int(porce/100 * pts1.shape[0])
else:
n_key = int(opt)
s_pts1 = pts1.copy()[:n_key,:]
s_pts2 = pts2.copy()[:n_key,:]
s_desc1 = desc1.copy().astype('float32')[:n_key,:]
s_desc2 = desc2.copy().astype('float32')[:n_key,:]
if 'orb' in args_opt:
s_desc1 = s_desc1.astype(np.uint8)
s_desc2 = s_desc2.astype(np.uint8)
if match == '2-cross':
if 'orb' in args_opt:
bf = cv2.BFMatcher(cv2.NORM_HAMMING, True)
else:
bf = cv2.BFMatcher(cv2.NORM_L2, True)
matches = bf.match(s_desc1, s_desc2)
elif match == 'ratio':
if 'orb' in args_opt:
bf = cv2.BFMatcher(cv2.NORM_HAMMING, False)
else:
bf = cv2.BFMatcher(cv2.NORM_L2, False)
matches = bf.knnMatch(s_desc1,s_desc2, k=2)
good = []
for m,n in matches:
if m.distance < 0.75*n.distance:
good.append(m)
matches = good
M = np.zeros((2,len(matches)))
for ind, match in zip(np.arange(len(matches)),matches):
M[0,ind] = match.queryIdx
M[1,ind] = match.trainIdx
num_M = M.shape[1]
return s_pts1, s_pts2, s_pts1[M[0,:].astype(int),:3], s_pts2[M[1,:].astype(int),:3]
def get_error(x1, x2, Rx, Tx):
S = computeEssentialMatrixByRANSAC(x1, x2)
I = S[1]
I = I.astype(np.int64)
x1 = x1[I,:]
x2 = x2[I,:]
F = calc_ematrix(x1,x2)
R1,R2,T1,T2 = decomposeE(F)
R,T = choose_rt(R1,R2,T1,T2,x1,x2)
R_error = r_error(Rx,R)
T_error = t_error(Tx,T)
return R_error, T_error
def get_descriptor(descriptor):
if descriptor == 'sphorb':
return 'sphorb', 'erp', 640
elif descriptor == 'sift':
return 'sift', 'erp', 512
elif descriptor == 'tsift':
return 'sift', 'tangent', 512
elif descriptor == 'orb':
return 'orb', 'erp', 512
elif descriptor == 'torb':
return 'orb', 'tangent', 512
elif descriptor == 'spoint':
return 'superpoint', 'erp', 512
elif descriptor == 'tspoint':
return 'superpoint', 'tangent', 512
def AUC(ROT, TRA, MET, L):
RAUC = np.zeros(len(L))
TAUC = np.zeros(len(L))
for index, t in enumerate(L):
ids = np.where(ROT<np.radians(t))[0]
RAUC[index] = len(ids)/len(ROT)
for index, t in enumerate(L):
ids = np.where(TRA<np.radians(t))[0]
TAUC[index] = len(ids)/len(TRA)
return RAUC, TAUC, np.array(MET)
def get_data(DATAS):
if len(DATAS) == 1:
data = DATAS[0]
elif set(['Urban1','Urban2','Urban3','Urban4']) == set(DATAS):
data = 'Outdoor'
elif set(['Realistic','Interior1','Interior2','Room','Classroom']) == set(DATAS):
data = 'Indoor'
elif set(['Urban1_R','Urban2_R','Urban3_R','Urban4_R','Realistic_R','Interior1_R','Interior2_R','Room_R','Classroom_R']) == set(DATAS):
data = 'OnlyRot'
elif set(['Urban1_T','Urban2_T','Urban3_T','Urban4_T','Realistic_T','Interior1_T','Interior2_T','Room_T','Classroom_T']) == set(DATAS):
data = 'OnlyTra'
else:
data = ''
for DA in DATAS:
data+=DA
return data
def get_kd(array):
array = np.array(array)
delimiter = int(array[-1])
A = array[:-1]
K = A[:delimiter].reshape(-1,3)
D = A[delimiter:].reshape(-1,32)
return K,D
def main():
parser = argparse.ArgumentParser(description = 'Tangent Plane')
parser.add_argument('--points', type=int, default = 12000)
parser.add_argument('--match', default="ratio")
parser.add_argument('--g_metrics',default="False")
parser.add_argument('--solver', default="None")
parser.add_argument('--inliers', default="8PA")
parser.add_argument('--datas' , nargs='+')
parser.add_argument('--descriptors', nargs='+')
args = parser.parse_args()
DATAS = args.datas
DESCRIPTORS = args.descriptors
# ----------------------------------------------
# Parameters
# ----------------------------------------------
NUM = 0
R_ERROR, T_ERROR = [], []
for i in range(len(DESCRIPTORS)):
R_ERROR.append([])
T_ERROR.append([])
if args.g_metrics == "False":
METRICS = np.zeros((len(DESCRIPTORS),2))
metrics = ['Matched','Keypoint']
else:
METRICS = np.zeros((len(DESCRIPTORS),7))
metrics = ['Matched','Keypoint','Pmr','Pr','R','Ms','E']
data = get_data(DATAS)
TIMES = []
for data in DATAS:
mypath = os.path.join('data',data)
paths = [os.path.join(os.getcwd(),'data',data,f) for f in listdir('data/'+data) if isdir(join(mypath, f))]
NUM = NUM + len(paths)
std = []
for path in tqdm(paths):
for indicador, descriptor in enumerate(DESCRIPTORS):
try:
opt, mode, sphered = get_descriptor(descriptor)
base_order = 1 # Base sphere resolution
sample_order = 8 # Determines sample resolution (10 = 2048 x 4096)
scale_factor = 1.0 # How much to scale input equirectangular image by
save_ply = False # Whether to save the PLY visualizations too
dim = np.array([2*sphered, sphered])
path_o = path + '/O.png'
path_r = path + '/R.png'
if opt != 'sphorb':
# ----------------------------------------------
# Compute necessary data
# ----------------------------------------------
# 80 baricenter points
corners = tangent_image_corners(base_order, sample_order)
pts1, desc1 = process_image_to_keypoints(path_o, corners, scale_factor, base_order, sample_order, opt, mode)
pts2, desc2 = process_image_to_keypoints(path_r, corners, scale_factor, base_order, sample_order, opt, mode)
pts1, pts2, desc1, desc2 = sort_key(pts1, pts2, desc1, desc2, args.points)
else:
#os.system('mogrify -format jpg '+path+'/*.png')
os.chdir('SPHORB-master/')
pts1, desc1 = get_kd(sphorb.sphorb(path_o, args.points))
pts2, desc2 = get_kd(sphorb.sphorb(path_r, args.points))
os.chdir('../')
depth = np.load("results/depth/"+data+".npy")
if len(pts1.shape) == 1:
pts1 = pts1.reshape(1,-1)
if len(pts2.shape) == 1:
pts2 = pts2.reshape(1,-1)
Rx = np.load(path+"/R.npy")
Tx = np.load(path+"/T.npy")
for indice, option in enumerate(['100p']):
if pts1.shape[0] > 0 or pts2.shape[0] >0:
s_pts1, s_pts2, x1, x2 = matched_points(pts1, pts2, desc1, desc2, option, opt, args.match)
z_d = depth[(x1[:,1]*512/sphered).astype('int')%512,(x1[:,0]*512/sphered).astype('int')%1024]
z_d2 = depth[(s_pts1[:,1]*512/sphered).astype('int')%512,(s_pts1[:,0]*512/sphered).astype('int')%1024]
x1,x2 = coord_3d(x1, dim), coord_3d(x2, dim)
x2_ = (x1.copy()*z_d.reshape(-1,1))@Rx.T + Tx
x2_ = x2_/np.linalg.norm(x2_,axis=1).reshape(-1,1)
s_pts1, s_pts2 = coord_3d(s_pts1, dim), coord_3d(s_pts2, dim)
s_pts2_ = (s_pts1.copy()*z_d2.reshape(-1,1))@Rx.T + Tx
s_pts2_ = s_pts2_/np.linalg.norm(s_pts2_,axis=1).reshape(-1,1)
if args.g_metrics == "True":
d2 = Tall_error(s_pts2, s_pts2_)
d = FT_error(x2,x2_)
EPS = 1e-5
pmr = x1.shape[0]/(s_pts1.shape[0]+EPS)
Pr = np.sum(d<0.1)/(x1.shape[0]+EPS)
r = np.sum(d<0.1)/(np.sum(d2<0.01)+EPS)
ms = np.sum(d<0.1)/(s_pts1.shape[0]+EPS)
En = calc_entropy(x1)
if x1.shape[0] < 8:
R_error, T_error = 3.14, 3.14
else:
inicio = time.time()
#time.sleep(1)
if args.solver == 'None':
E, cam = get_cam_pose_by_ransac_8pa(x1.copy().T,x2.copy().T, get_E = True, I = args.inliers)
elif args.solver == 'SK':
E, can = get_cam_pose_by_ransac_opt_SK(x1.copy().T,x2.copy().T, get_E = True, I = args.inliers)
elif args.solver == 'GSM':
E, can = get_cam_pose_by_ransac_GSM(x1.copy().T,x2.copy().T, get_E = True, I = args.inliers)
elif args.solver == 'GSM_wRT':
E, can = get_cam_pose_by_ransac_GSM_const_wRT(x1.copy().T,x2.copy().T, get_E = True, I = args.inliers)
elif args.solver == 'GSM_SK':
E, can = get_cam_pose_by_ransac_GSM_const_wSK(x1.copy().T,x2.copy().T, get_E = True, I = args.inliers)
fin = time.time()
TIMES.append(fin-inicio)
R1_,R2_,T1_,T2_ = decomposeE(E.T)
R_,T_ = choose_rt(R1_,R2_,T1_,T2_,x1,x2)
#R_error, T_error = get_error(x1, x2, Rx, Tx)
R_error, T_error = r_error(Rx,R_), t_error(Tx,T_)
R_ERROR[indicador].append(R_error)
T_ERROR[indicador].append(T_error)
if args.g_metrics == "False":
METRICS[indicador,:] = METRICS[indicador,:] + [x1.shape[0], (s_pts1.shape[0]+s_pts2.shape[1])/2]
else:
METRICS[indicador,:] = METRICS[indicador,:] + [x1.shape[0], (s_pts1.shape[0]+s_pts2.shape[1])/2,pmr,Pr,r,ms,En]
std.append(x1.shape[0])
except:
print("Unexpected error")
#print(np.array(std).std())
print(np.mean(np.array(TIMES)))
print('ALL:')
print(np.mean(np.array(TIMES)))
N_ARRAY = np.zeros((len(DESCRIPTORS),1))
for indice, _ in enumerate(DESCRIPTORS):
N_ARRAY[indice,0] = len(R_ERROR[indice])
METRICS = METRICS/N_ARRAY
METRICS = METRICS*100
for file in ['100p']:
data = get_data(DATAS)
print("Results in data: "+ data +" with " + file + " of keypoints")
P = DESCRIPTORS
L = [0.1, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0, 20.0]
ROT, TRA, MET = np.zeros((len(P),len(L))), np.zeros((len(P),len(L))), np.zeros((len(P),METRICS.shape[1]))
for i in range(len(P)):
ROT[i,:], TRA[i,:], MET[i,:] = AUC(R_ERROR[i], T_ERROR[i], METRICS[i], L)
mat_R = {}
mat_T = {}
mat_M = {}
for ind, l in enumerate(L):
mat_R[l]=ROT[:,ind]*100
mat_T[l]=TRA[:,ind]*100
for ind, m in enumerate(metrics):
mat_M[m]=MET[:,ind]*100
ROTATION = pd.DataFrame(mat_R, index=P)
os.system('mkdir -p '+'results/metrics/'+data+'_'+file+'_'+args.inliers+'_'+args.solver)
ROTATION.to_csv('results/metrics/'+data+'_'+file+'_'+args.inliers+'_'+args.solver+'/rotation.txt',float_format='%.2f')
TRANSLATION = pd.DataFrame(mat_T, index=P)
TRANSLATION.to_csv('results/metrics/'+data+'_'+file+'_'+args.inliers+'_'+args.solver+'/translation.txt', float_format='%.2f')
METRICS = pd.DataFrame(mat_M, index=P)
METRICS.to_csv('results/metrics/'+data+'_'+file+'_'+args.inliers+'_'+args.solver+'/metrics.txt', float_format='%.3f')
print('ROTATION')
print(ROT)
print('TRANSLATION')
print(TRA)
print('GENERIC METRICS')
print(MET)
# fig, ax = plt.subplots(1,figsize=(8,6))
#
# for i in range(len(P)):
# ax.plot(L,ROT[i,:],'-o',linewidth=2,label = P[i])
#
# plt.xlabel('degree')
# plt.ylabel('accuracy')
# ax.legend(loc="lower right")
#
# plt.savefig('results/metrics/'+data+'_'+file+'_'+args.inliers+'_'+args.solver+'/Rotation.png')
#
# plt.clf()
#
#
# fig, ax = plt.subplots(1,figsize=(8,6))
#
# for i in range(len(P)):
# ax.plot(L,TRA[i,:],'-o',linewidth=2,label = P[i])
#
# plt.xlabel('degree')
# plt.ylabel('accuracy')
# ax.legend(loc="lower right")
#
# plt.savefig('results/metrics/'+data+'_'+file+'_'+args.inliers+'_'+args.solver+'/Translation.png')
#
if __name__ == '__main__':
main()