/
non_learning_completion_multi_res.py
242 lines (161 loc) · 7.83 KB
/
non_learning_completion_multi_res.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
from data_read import *
import matplotlib.pyplot as plt
import cv2
import numpy as np
from tools import *
from evaluation import *
import time
import tensorflow as tf
def do_range_projection_try(points,proj_H=64,proj_W=2048,fov_up=3.0,fov_down=-18.0):
proj_range = np.full((proj_H, proj_W), -1,dtype=np.float32)
# unprojected range (list of depths for each point)
unproj_range = np.zeros((0, 1), dtype=np.float32)
# projected point cloud xyz - [H,W,3] xyz coord (-1 is no data)
proj_xyz = np.full((proj_H, proj_W, 3), -1,dtype=np.float32)
# projected index (for each pixel, what I am in the pointcloud)
# [H,W] index (-1 is no data)
proj_idx = np.full((proj_H, proj_W), -1,
dtype=np.int32)
# for each point, where it is in the range image
proj_x = np.zeros((0, 1), dtype=np.float32) # [m, 1]: x
proj_y = np.zeros((0, 1), dtype=np.float32) # [m, 1]: y
# mask containing for each pixel, if it contains a point or not
proj_mask = np.zeros((proj_H, proj_W),dtype=np.int32)
# laser parameters
fov_up = fov_up / 180.0 * np.pi # field of view up in rad
fov_down = fov_down / 180.0 * np.pi # field of view down in rad
fov = abs(fov_down) + abs(fov_up) # get field of view total in rad
# get depth of all points
depth = np.linalg.norm(points, 2, axis=1)
# get scan components
scan_x = points[:, 0]
scan_y = points[:, 1]
scan_z = points[:, 2]
# get angles of all points
yaw = -np.arctan2(scan_y, scan_x)
pitch = np.arcsin(scan_z / depth)
# get projections in image coords
proj_x = 0.5 * (yaw / np.pi + 1.0) # in [0.0, 1.0]
proj_y = 1.0 - (pitch + abs(fov_down)) / fov # in [0.0, 1.0]
return proj_x,proj_y
def outlier_removal_mask(lidar,line_num=64,height_offset=100):
height,width=np.shape(np.squeeze(lidar))
height_bin=np.round((height-height_offset)/line_num-1)
width_bin=np.round(width*line_num/(np.sum(lidar>0.1))-1)
total_points,x_indices,y_indices,width_image,height_image=get_all_points(lidar,intrinsic)
proj_x,proj_y=do_range_projection_try(total_points)
project_x=np.zeros((height,width))
project_y=np.zeros((height,width))
project_x[y_indices,x_indices]=proj_x
project_y[y_indices,x_indices]=proj_y
lidar_pre=np.expand_dims(np.squeeze(lidar),axis=0)
lidar_pre=np.expand_dims(lidar_pre,axis=-1)
lidar_trunck=tf.image.extract_patches(images=lidar_pre,sizes=[1, height_bin*2+1, width_bin*2+1, 1],strides=[1, 1, 1, 1],rates=[1, 1, 1, 1],padding='SAME')
expand_size=(height_bin*2+1)*(width_bin*2+1)
expand_size=np.int32(expand_size)
lidar_pre_expand=tf.tile(lidar_pre.astype(np.float32),[1,1,1,expand_size])
project_x_pre=np.expand_dims(np.squeeze(project_x),axis=0)
project_x_pre=np.expand_dims(project_x_pre,axis=-1)
project_x_trunck=tf.image.extract_patches(images=project_x_pre,sizes=[1, height_bin*2+1, width_bin*2+1, 1],strides=[1, 1, 1, 1],rates=[1, 1, 1, 1],padding='SAME')
project_y_pre=np.expand_dims(np.squeeze(project_y),axis=0)
project_y_pre=np.expand_dims(project_y_pre,axis=-1)
project_y_trunck=tf.image.extract_patches(images=project_y_pre,sizes=[1, height_bin*2+1, width_bin*2+1, 1],strides=[1, 1, 1, 1],rates=[1, 1, 1, 1],padding='SAME')
height_image_pre=np.expand_dims(np.squeeze(height_image).astype(np.double),axis=0)
height_image_pre=np.expand_dims(height_image_pre,axis=-1)
height_image_trunck=tf.image.extract_patches(images=height_image_pre,sizes=[1, height_bin*2+1, width_bin*2+1, 1],strides=[1, 1, 1, 1],rates=[1, 1, 1, 1],padding='SAME')
width_image_pre=np.expand_dims(np.squeeze(width_image).astype(np.double),axis=0)
width_image_pre=np.expand_dims(width_image_pre,axis=-1)
width_image_trunck=tf.image.extract_patches(images=width_image_pre,sizes=[1, height_bin*2+1, width_bin*2+1, 1],strides=[1, 1, 1, 1],rates=[1, 1, 1, 1],padding='SAME')
lidar_residual=lidar_pre-lidar_trunck
project_x_residual=project_x_pre-project_x_trunck
project_y_residual=project_y_pre-project_y_trunck
height_image_residual=height_image_pre-height_image_trunck
width_image_residual=width_image_pre-width_image_trunck
zero_mask=np.logical_and(lidar_pre_expand>0.1,lidar_trunck>0.1)
x_mask_1=np.logical_and(project_x_residual>0.0000,width_image_residual<=0)
x_mask_2=np.logical_and(project_x_residual<0.0000,width_image_residual>=0)
x_mask=np.logical_or(x_mask_1,x_mask_2)
x_mask=np.logical_and(x_mask,zero_mask)
y_mask_1=np.logical_and(project_y_residual>0,height_image_residual<=0)
y_mask_2=np.logical_and(project_y_residual<0,height_image_residual>=0)
y_mask=np.logical_or(y_mask_1,y_mask_2)
y_mask=np.logical_and(y_mask,zero_mask)
lidar_mask=np.logical_and(lidar_residual>3.0,lidar_pre>0.01)
final_mask=np.logical_and(lidar_mask,np.logical_or(x_mask,y_mask))
final_mask=np.squeeze(final_mask)
final_mask=np.sum(final_mask,axis=-1)
final_mask=np.expand_dims(final_mask>0,axis=0)
return final_mask
evaluate=Result()
rmse_total=0
mae_total=0
irmse_total=0
imae_total=0
#validation
A=calculate_normal()
# can be 32 or 16
line_number=64
kernelx=self_gaussian(kernel_size=7,g_range=2.5)
time_interval=[]
threshold=0.1
for i in range(1000):
print (i)
img,lidar,gt,intrinsic= read_one_val(i,line_number=64)
# prepare camera parameters
px,py,fx,fy=construct_px_py_fx_fy(lidar,intrinsic)
# remove outliers
if i>20:
time_a=time.time()
outpier_mask=outlier_removal_mask(lidar,line_num=64,height_offset=100)
lidar_new=(1.0-outpier_mask)*lidar
# get all points
total_points,x_indices,y_indices,width_image,height_image=get_all_points(lidar_new,intrinsic)
# map points on range image
range_image,proj_xyz,proj_idx,proj_mask=do_range_projection(total_points)
# fill in range image
range_image_new=fill_spherical(range_image)
# calculate normal in spherical
normal_spherical=A.calculate_normal(range_image_new)
# put normal from spherical on original image
normal_img=put_normal_on_image(normal_spherical,x_indices,y_indices,proj_idx,proj_mask)
a=normal_img[:,:,0]
b=normal_img[:,:,1]
c=normal_img[:,:,2]
depth_map,a_list_all,b_list_all,c_list_all,height_list_offset,width_list_offset=Distance_Transform(lidar_new,a,b,c,width_image,height_image)
upper=(width_list_offset/fx*a_list_all+height_list_offset/fy*b_list_all)
lower=(width_image-px)/fx*a_list_all+(height_image-py)/fy*b_list_all+c_list_all
lower[np.logical_and(np.abs(lower)<threshold,lower>0)]=threshold
lower[np.logical_and(np.abs(lower)<threshold,lower<0)]=-threshold
residual=upper/lower
depth_predicted=depth_map+residual*depth_map
depth_predicted = cv2.filter2D(depth_predicted, -1, kernelx)
depth_predicted=np.squeeze(lidar_new)+depth_predicted*(1.0-np.squeeze(lidar_new)>0.01)
depth_predicted[depth_predicted<0.9]=0.9
depth_predicted[depth_predicted>80.0]=80.0
if i>20:
time_b=time.time()
time_interval.append(time_b-time_a)
evaluate.evaluate(np.squeeze(depth_predicted),np.squeeze(gt))
irmse, imae, mse, rmse, mae=evaluate.irmse, evaluate.imae, evaluate.mse, evaluate.rmse, evaluate.mae
rmse_total+=rmse
mae_total+=mae
irmse_total+=irmse
imae_total+=imae
print ("rmse:")
print (rmse_total/(i+1))
print ("mae:")
print (mae_total/(1+i))
print ("irmse:")
print (irmse_total/(1+i))
print ("imae:")
print (imae_total/(1+i))
print ("rmse:")
print (rmse_total/1000.0)
print ("mae:")
print (mae_total/1000.0)
print ("irmse:")
print (irmse_total/1000.0)
print ("imae:")
print (imae_total/1000.0)
print ('method time')
print (np.sum(time_interval)/len(time_interval))