-
Notifications
You must be signed in to change notification settings - Fork 0
/
SFEGO_3D.py
341 lines (292 loc) · 14.9 KB
/
SFEGO_3D.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
import numpy as np
import math
import pyopencl as cl
import cv2
import skimage
import time
M_PI=3.14159265358979323846
Platfrom_ID=0
Device_ID=0
def build_list_3d_sphere(radius):
ar_len=0
x_list=[]
y_list=[]
z_list=[]
unit_x_list=[]
unit_y_list=[]
unit_z_list=[]
theta_list=[]
phi_list=[]
radius_list=[]
for i in range(-radius,radius+1):
for j in range(-radius,radius+1):
for k in range(-radius,radius+1):
r=np.sqrt(i*i+j*j+k*k)
if ((r < radius+1.0) and not (i==0 and j==0 and k==0)):
x_list.append(i)
y_list.append(j)
z_list.append(k)
unit_x_list.append(i/r)
unit_y_list.append(j/r)
unit_z_list.append(k/r)
theta=math.atan2(i,j)
if theta<0.0:
theta+=M_PI*2
theta_list.append(theta)
phi=math.atan2(k,np.sqrt(i*i+j*j))
if phi<0.0:
phi+=M_PI*2
phi_list.append(phi)
radius_list.append(r)
zipped=zip(x_list, y_list, z_list, unit_x_list, unit_y_list, unit_z_list, phi_list, theta_list, radius_list)
# 0 1 2 3 4 5 6 7 8
zipped=sorted(zipped, key = lambda x: (x[6], x[7], x[8]))
#zipped=zip(x_list, y_list, deg_list, radius_list)
#zipped=sorted(zipped, key = lambda x: (x[2], x[3]))
return zipped
def generate_surface_dp_list(ar_list, radius):
x_list, y_list, z_list, unit_x_list, unit_y_list, unit_z_list, phi_list, theta_list, radius_list = zip(*ar_list)
# find sphere surface index in the list
surface_indexs=[]
for index in range(len(radius_list)):
min_radius=radius-0.0001
if radius_list[index]>min_radius:
surface_indexs.append(index)
hemisphere_indexs=[] #len should equalt to len(surface_indexs)
for idx in range(len(surface_indexs)):
center_index=surface_indexs[idx]
center_unit_x=unit_x_list[center_index]
center_unit_y=unit_y_list[center_index]
center_unit_z=unit_z_list[center_index]
positive_hemisphere=[]
negative_hemisphere=[]
for index in range(len(radius_list)):
if center_index == index:
continue
target_unit_x=unit_x_list[index]
target_unit_y=unit_y_list[index]
target_unit_z=unit_z_list[index]
target_dot_prdouct=(center_unit_x*target_unit_x)+(center_unit_y*target_unit_y)+(center_unit_z*target_unit_z)
if target_dot_prdouct>0.0001:
positive_hemisphere.append(index)
elif target_dot_prdouct<-0.0001:
negative_hemisphere.append(index)
hemisphere_indexs.append([positive_hemisphere, negative_hemisphere])
#print(len(hemisphere_indexs))
hemisphere_dp_pos_add=[]
hemisphere_dp_pos_sub=[]
hemisphere_dp_neg_add=[]
hemisphere_dp_neg_sub=[]
# First add all
hemisphere_dp_pos_add.append(hemisphere_indexs[0][0])
hemisphere_dp_pos_sub.append([])
hemisphere_dp_neg_add.append(hemisphere_indexs[0][1])
hemisphere_dp_neg_sub.append([])
prev_pos=hemisphere_indexs[0][0]
prev_neg=hemisphere_indexs[0][1]
for idx in range(1,len(surface_indexs)):
current_pos=hemisphere_indexs[idx][0]
current_neg=hemisphere_indexs[idx][1]
pos_add=list(set(current_pos)-set(prev_pos))
pos_sub=list(set(prev_pos)-set(current_pos))
neg_add=list(set(current_neg)-set(prev_neg))
neg_sub=list(set(prev_neg)-set(current_neg))
hemisphere_dp_pos_add.append(pos_add)
hemisphere_dp_pos_sub.append(pos_sub)
hemisphere_dp_neg_add.append(neg_add)
hemisphere_dp_neg_sub.append(neg_sub)
prev_pos=current_pos
prev_neg=current_neg
dp_pos_add_start_idx=0
dp_pos_add_start_idxs=[]
dp_pos_add_start_lens=[]
dp_pos_sub_start_idx=0
dp_pos_sub_start_idxs=[]
dp_pos_sub_start_lens=[]
dp_neg_add_start_idx=0
dp_neg_add_start_idxs=[]
dp_neg_add_start_lens=[]
dp_neg_sub_start_idx=0
dp_neg_sub_start_idxs=[]
dp_neg_sub_start_lens=[]
for idx in range(0, len(surface_indexs)):
dp_pos_add_start_idxs.append(dp_pos_add_start_idx)
dp_pos_sub_start_idxs.append(dp_pos_sub_start_idx)
dp_neg_add_start_idxs.append(dp_neg_add_start_idx)
dp_neg_sub_start_idxs.append(dp_neg_sub_start_idx)
dp_pos_add_start_len=len(hemisphere_dp_pos_add[idx])
dp_pos_add_start_lens.append(dp_pos_add_start_len)
dp_pos_add_start_idx+=dp_pos_add_start_len
dp_pos_sub_start_len=len(hemisphere_dp_pos_sub[idx])
dp_pos_sub_start_lens.append(dp_pos_sub_start_len)
dp_pos_sub_start_idx+=dp_pos_sub_start_len
dp_neg_add_start_len=len(hemisphere_dp_neg_add[idx])
dp_neg_add_start_lens.append(dp_neg_add_start_len)
dp_neg_add_start_idx+=dp_neg_add_start_len
dp_neg_sub_start_len=len(hemisphere_dp_neg_sub[idx])
dp_neg_sub_start_lens.append(dp_neg_sub_start_len)
dp_neg_sub_start_idx+=dp_neg_sub_start_len
hemisphere_dp_pos_add=sum(hemisphere_dp_pos_add, [])
hemisphere_dp_pos_sub=sum(hemisphere_dp_pos_sub, [])
hemisphere_dp_neg_add=sum(hemisphere_dp_neg_add, [])
hemisphere_dp_neg_sub=sum(hemisphere_dp_neg_sub, [])
return surface_indexs, \
hemisphere_dp_pos_add, hemisphere_dp_pos_sub, hemisphere_dp_neg_add, hemisphere_dp_neg_sub, \
dp_pos_add_start_idxs, dp_pos_sub_start_idxs, dp_neg_add_start_idxs, dp_neg_sub_start_idxs, \
dp_pos_add_start_lens, dp_pos_sub_start_lens, dp_neg_add_start_lens, dp_neg_sub_start_lens
#Initial PyOpenCL
#ctx = cl.create_some_context(interactive=False)
print("Build OpenCL Kernel...")
ctx = cl.Context([cl.get_platforms()[Platfrom_ID].get_devices()[Device_ID]])
queue = cl.CommandQueue(ctx)
prg = cl.Program(ctx, open('kernel_3d.cl').read()).build()
knl_gradient = prg.SFEGO_3d_gradient
knl_integral = prg.SFEGO_3d_integral
mf = cl.mem_flags
print("Done!!\n")
def SFEGO_3D(np_input_data, dim_x, dim_y, dim_z, radius):
ar_list = build_list_3d_sphere(radius)
x_list, y_list, z_list, unit_x_list, unit_y_list, unit_z_list, phi_list, theta_list, radius_list = zip(*ar_list)
list_len=len(x_list)
surface_indexs, \
hemisphere_dp_pos_add, hemisphere_dp_pos_sub, hemisphere_dp_neg_add, hemisphere_dp_neg_sub, \
dp_pos_add_start_idxs, dp_pos_sub_start_idxs, dp_neg_add_start_idxs, dp_neg_sub_start_idxs, \
dp_pos_add_start_lens, dp_pos_sub_start_lens, dp_neg_add_start_lens, dp_neg_sub_start_lens = generate_surface_dp_list(ar_list, radius)
dp_len=len(surface_indexs)
np_x_list = np.asarray(x_list).astype(np.int32)
np_y_list = np.asarray(y_list).astype(np.int32)
np_z_list = np.asarray(z_list).astype(np.int32)
np_unit_x_list = np.asarray(unit_x_list).astype(np.float32)
np_unit_y_list = np.asarray(unit_y_list).astype(np.float32)
np_unit_z_list = np.asarray(unit_z_list).astype(np.float32)
np_surface_indexs = np.asarray(surface_indexs).astype(np.int32)
np_hemisphere_dp_pos_add = np.asarray(hemisphere_dp_pos_add).astype(np.int32)
np_hemisphere_dp_pos_sub = np.asarray(hemisphere_dp_pos_sub).astype(np.int32)
np_hemisphere_dp_neg_add = np.asarray(hemisphere_dp_neg_add).astype(np.int32)
np_hemisphere_dp_neg_sub = np.asarray(hemisphere_dp_neg_sub).astype(np.int32)
np_dp_pos_add_start_idxs = np.asarray(dp_pos_add_start_idxs).astype(np.int32)
np_dp_pos_sub_start_idxs = np.asarray(dp_pos_sub_start_idxs).astype(np.int32)
np_dp_neg_add_start_idxs = np.asarray(dp_neg_add_start_idxs).astype(np.int32)
np_dp_neg_sub_start_idxs = np.asarray(dp_neg_sub_start_idxs).astype(np.int32)
np_dp_pos_add_start_lens = np.asarray(dp_pos_add_start_lens).astype(np.int32)
np_dp_pos_sub_start_lens = np.asarray(dp_pos_sub_start_lens).astype(np.int32)
np_dp_neg_add_start_lens = np.asarray(dp_neg_add_start_lens).astype(np.int32)
np_dp_neg_sub_start_lens = np.asarray(dp_neg_sub_start_lens).astype(np.int32)
data = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_input_data)
diff = cl.Buffer(ctx, mf.READ_WRITE, np_input_data.nbytes)
direct_x = cl.Buffer(ctx, mf.READ_WRITE, np_input_data.nbytes)
direct_y = cl.Buffer(ctx, mf.READ_WRITE, np_input_data.nbytes)
direct_z = cl.Buffer(ctx, mf.READ_WRITE, np_input_data.nbytes)
result = cl.Buffer(ctx, mf.READ_WRITE, np_input_data.nbytes)
list_x = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_x_list)
list_y = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_y_list)
list_z = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_z_list)
list_unit_x = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_unit_x_list)
list_unit_y = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_unit_y_list)
list_unit_z = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_unit_z_list)
cl_surface_indexs = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_surface_indexs)
cl_hemisphere_dp_pos_add = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_hemisphere_dp_pos_add)
cl_hemisphere_dp_pos_sub = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_hemisphere_dp_pos_sub)
cl_hemisphere_dp_neg_add = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_hemisphere_dp_neg_add)
cl_hemisphere_dp_neg_sub = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_hemisphere_dp_neg_sub)
cl_dp_pos_add_start_idxs = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_pos_add_start_idxs)
cl_dp_pos_sub_start_idxs = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_pos_sub_start_idxs)
cl_dp_neg_add_start_idxs = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_neg_add_start_idxs)
cl_dp_neg_sub_start_idxs = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_neg_sub_start_idxs)
cl_dp_pos_add_start_lens = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_pos_add_start_lens)
cl_dp_pos_sub_start_lens = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_pos_sub_start_lens)
cl_dp_neg_add_start_lens = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_neg_add_start_lens)
cl_dp_neg_sub_start_lens = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=np_dp_neg_sub_start_lens)
knl_gradient(queue, (dim_x, dim_y, dim_z), None, data, diff, direct_x, direct_y, direct_z, \
list_x, list_y, list_z, list_unit_x, list_unit_y, list_unit_z, \
cl_surface_indexs, \
cl_hemisphere_dp_pos_add, cl_hemisphere_dp_pos_sub, \
cl_hemisphere_dp_neg_add, cl_hemisphere_dp_neg_sub, \
cl_dp_pos_add_start_idxs, cl_dp_pos_sub_start_idxs, \
cl_dp_neg_add_start_idxs, cl_dp_neg_sub_start_idxs, \
cl_dp_pos_add_start_lens, cl_dp_pos_sub_start_lens, \
cl_dp_neg_add_start_lens, cl_dp_neg_sub_start_lens, \
np.int32(list_len), np.int32(dp_len), \
np.int32(dim_x), np.int32(dim_y), np.int32(dim_z))
knl_integral(queue, (dim_x, dim_y, dim_z), None, result, diff, direct_x, direct_y, direct_z, \
list_x, list_y, list_z, list_unit_x, list_unit_y, list_unit_z, \
np.int32(list_len), np.int32(dim_x), np.int32(dim_y), np.int32(dim_z))
np_result = np.empty_like(np_input_data)
cl.enqueue_copy(queue, np_result, result)
np_result = np_result / list_len
data.release()
diff.release()
direct_x.release()
direct_y.release()
direct_z.release()
result.release()
list_x.release()
list_y.release()
list_z.release()
list_unit_x.release()
list_unit_y.release()
list_unit_z.release()
cl_surface_indexs.release()
cl_hemisphere_dp_pos_add.release()
cl_hemisphere_dp_pos_sub.release()
cl_hemisphere_dp_neg_add.release()
cl_hemisphere_dp_neg_sub.release()
cl_dp_pos_add_start_idxs.release()
cl_dp_pos_sub_start_idxs.release()
cl_dp_neg_add_start_idxs.release()
cl_dp_neg_sub_start_idxs.release()
cl_dp_pos_add_start_lens.release()
cl_dp_pos_sub_start_lens.release()
cl_dp_neg_add_start_lens.release()
cl_dp_neg_sub_start_lens.release()
return np_result
def GenereateSimulationData():
dim_x=128
dim_y=129
dim_z=130
np_data=np.zeros(dim_x*dim_y*dim_z).astype(np.float32)
for x in range(dim_x):
for y in range(dim_y):
for z in range(dim_z):
cx=64-x
cy=64-y
cz=64-z
radius=np.sqrt(cx*cx+cy*cy+cz*cz)
np_data[z*dim_y*dim_x+y*dim_x+x]+=math.sin(radius/3.0)
np_data[z*dim_y*dim_x+y*dim_x+x]+=math.sin(x)+math.sin(y)+math.sin(z)
return np_data, dim_x, dim_y, dim_z
print("Generate Simulation Data...")
np_data, dim_x, dim_y, dim_z=GenereateSimulationData()
print("Done!!\n")
start_time = time.time()
file = open('default_radius')
for line in file:
fields = line.strip().split()
resize_ratio=float(fields[0])
execute_radius=int(fields[1])
target_dim_x=int(dim_x/resize_ratio)
target_dim_y=int(dim_y/resize_ratio)
target_dim_z=int(dim_z/resize_ratio)
print("resize_ratio="+str(resize_ratio)+" execute_radius="+str(execute_radius)+" effective_radius="+str(resize_ratio*execute_radius)+" size:(z,y,x)=("+str(target_dim_z)+", "+str(target_dim_y)+", "+str(target_dim_x)+")")
np_3d_data = np_data.reshape((dim_z, dim_y, dim_x))
np_3d_input_data = skimage.transform.resize(np_3d_data, (target_dim_z, target_dim_y, target_dim_x))
np_input_data = np_3d_input_data.flatten()
np_result=SFEGO_3D(np_input_data, target_dim_x, target_dim_y, target_dim_z, execute_radius)
np_3d_result = np_result.reshape((target_dim_z, target_dim_y, target_dim_x))
np_3d_output_result = skimage.transform.resize(np_3d_result, (dim_z, dim_y, dim_x))
print("Done!!\n")
for z in range(dim_z):
np_2d_input = np_3d_data[z].copy()
np_2d_result = np_3d_output_result[z].copy()
result_min=np.min(np_2d_result)
result_max=np.max(np_2d_result)
np_2d_result = (255*(np_2d_result-result_min)/(result_max-result_min)).astype(np.uint8)
input_min=np.min(np_2d_input)
input_max=np.max(np_2d_input)
np_2d_input = (255*(np_2d_input-input_min)/(input_max-input_min)).astype(np.uint8)
output = cv2.hconcat([np_2d_input, np_2d_result])
cv2.imshow('Input v.s. Output', output)
cv2.waitKey(16)
end_time=time.time()
used_time=end_time-start_time
print("Used Time:", used_time)