-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathinference.py
490 lines (439 loc) · 20.8 KB
/
inference.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
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
import os, time, logging, copy, cv2, rospy
import numpy as np
from PIL import Image
from numpy import pi
from os.path import join
from scipy.spatial.transform import Rotation
import torch.nn as nn
import torch
from torchvision import models
from torchvision import transforms as T
from math import pi as pp
from math import cos, sin
from std_msgs.msg import Int64, Float64MultiArray
from franka_msgs.msg import FrankaState
class ArtificialSonographer:
def __init__(self, camera_id: int):
# initiate video stream
self.flag_stream = False
self.cap = None
self.crop_coord = (460, 140, 1200, 750)
self.__init_stream__(camera_id)
# initiate deep model
self.device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
self.model_list = []
self.trans = T.Compose([
T.Resize((224, 224)), T.ToTensor(),
T.Normalize([0.193, 0.193, 0.193], [0.224, 0.224, 0.224])
])
# read probe angle, ee_position exists when flag_position=True
self.flag_angle = False
self.ee_position = None
self.flag_position = False
self.angle_dict = {(0, 1): '4', (0, -1): '6',
(1, 1): '5', (1, -1): '8',
(2, 1): '7', (2, -1): '9', }
self.flag_contact = False
# initiate inference process and send actions
self.flag_scan = True # when robotic arm holds still
action_list = ['u', 'o', 'i', 'k', 'l', 'j',
'7', '9', '4', '6', '5', '8',
's', 'f']
action_name = ['lift-up', 'press-down', 'move-up', 'move-down',
'move-left', 'move-right', 'roll-anti', 'roll',
'yaw-left', 'yaw-right', 'pitch-down', 'pitch-up',
'start', 'finished']
self.action_dict = dict(zip(action_list, action_name))
# self.action_ref = ['k', 'l', '6', '5', '7', 'f0', 'f1', 'f']
self.action_ref = ['k', 'l', 'i', 'j', 'f']
# stage
self.flag_infrn = False # run inference and publish actions when True
self.flag_autoinf = False
self.stage_ind = 0 # state index {0: state1, 1: stage2}
self.flag_vote = False
# action control
self.pub = None
self.pub2 = None
self.signal_len = 100
self.signal_list = []
self.action = ord('s')
self.stride = 1000
self.action_count = 0
self.flag_move = False
self.flag_moving = False
self.flag_force = False
self.ee_force = None
self.stop_signal = 0
def __load_model__(self, model_path, num_class):
model = models.resnet34()
model.fc = nn.Sequential(
nn.Linear(model.fc.in_features, 256),
nn.Dropout(0.4),
nn.Linear(256, num_class))
model.load_state_dict(torch.load(model_path))
model.to(self.device)
model.eval()
return model
def __init_stream__(self, camera_id):
cap = cv2.VideoCapture(camera_id)
for _ in range(3):
_, _ = cap.read()
assert cap.isOpened(), f'\n{"=" * 69}\n *** ERROR: Invalid video stream. ' \
f'Check camera_id (0, 1, 2 or 3) or hws driver ***\n{"=" * 69}\n'
self.cap = cap
def streaming(self, flag, position):
ret, frame = self.cap.read()
if not ret:
self.cap.release()
raise Exception('** Video stream terminated **')
angles = [0, 0, 0]
if flag: # flag_position
angles, _ = probe_angle(position)
frame_show = frame[self.crop_coord[1]:self.crop_coord[3],
self.crop_coord[0]:self.crop_coord[2], :]
return frame_show, angles
def pred2action(self, vote_list):
action = None
for ind in range(len(vote_list)):
stride = abs(vote_list[ind])
if stride > 1:
action = self.action_ref[self.stage_ind][ind]
break
if action is None:
stride = 0
action = 'f'
return int(stride) + 1000, ord(action)
def Franka_cb(self, data):
# read end effector position
self.ee_position = data.O_T_EE
self.flag_position = True
# read end effector force
self.ee_force = data.K_F_ext_hat_K
# self.logger.info(f'ee force: {self.ee_force}')
def keyboard(self, data):
cur_key = chr((data.data)).lower()
# echo save switch
if cur_key == 'c':
self.flag_stream = not self.flag_stream
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current state of [Camera Switch]: '
f'{self.flag_stream:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
elif cur_key == 's':
self.flag_infrn = not self.flag_infrn
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current state of [Process Switch]: '
f'{self.flag_infrn:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
if self.flag_infrn:
self.logger.info(f'{"-" * 64}\n {">> Start model inference <<":^60} \n'
f'{"-" * 64}\n')
else:
self.logger.info(f'{"-" * 64}\n {"<< Model inference paused >>":^60} \n'
f'{"-" * 64}\n')
elif cur_key == 'a':
self.flag_angle = not self.flag_angle
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current state of [Angle Switch]: '
f'{self.flag_angle:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
elif cur_key == 'z':
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<19}* Launch probe to chest *'
f'{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
elif cur_key == 'b':
self.flag_contact = not self.flag_contact
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current state of [Contact Switch]: '
f'{self.flag_contact:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
elif cur_key == 'r':
self.stride = 1000 + 10
self.pub.publish(self.stride)
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current stride: '
f'{self.stride - 1000:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
elif cur_key == 'e':
self.stride = 1000 + 2
self.pub.publish(self.stride)
self.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current stride: '
f'{self.stride - 1000:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
elif cur_key == 'v': # auto inference
self.logger.info(
f'{"-" * 64}\n {">> Start auto inference <<":^60} \n{"-" * 64}\n')
self.flag_autoinf = True
self.stop_signal = 0
else:
pass
def action_control(self, msg):
'''
> Indicating action finish when receiving a full list of 1
with length 'self.signal_len'
> change self.signal_len for optimal performance
'''
self.signal_list.append(msg.data)
self.logger.debug(f'* signal data: {msg.data} *')
if len(self.signal_list) > self.signal_len:
self.signal_list = self.signal_list[-self.signal_len:]
if self.flag_move and not self.stride == 1000:
self.flag_move = False
action = chr(self.action)
self.pub.publish(self.stride)
self.pub.publish(self.action)
self.signal_list = []
self.flag_moving = True
action_name = self.action_dict[action]
self.logger.info(f'\t=> #{self.action_count:03d}-{action} '
f'stride-{self.stride - 1000} | "{action_name}"')
if not self.flag_contact:
os.system(f'spd-say "{action_name}" ')
if self.flag_moving and self.signal_list == [1] * self.signal_len:
self.flag_moving = False
self.flag_scan = True
action_name = self.action_dict[chr(self.action)]
self.logger.info(f'\t=> Action "{action_name}" complete \n{"-" * 48}')
os.system(f'spd-say "done"')
def init_log(self, log_root):
# dir and name
_, script_name = os.path.split(__file__)
script_name_aff, _ = script_name.split('.')
log_dir = join(log_root, script_name_aff)
os.makedirs(log_dir, exist_ok=True)
log_time = time.strftime('%Y%m%d_%H%M', time.localtime())
log_path = join(log_dir, f'{log_time}.log')
# config logger
file_handler = logging.FileHandler(log_path, encoding="UTF-8") # stdout to file
console_handler = logging.StreamHandler() # stdout to console
# set level
file_handler.setLevel('DEBUG')
console_handler.setLevel('INFO')
# output format
log_fmt = '%(message)s' # log_fmt = '[%(asctime)s] - <%(funcName)s> : %(levelname)s - %(message)s'
log_fmt_file = '[%(asctime)s] - %(message)s'
formatter = logging.Formatter(log_fmt)
formatter_file = logging.Formatter(log_fmt_file)
file_handler.setFormatter(formatter_file)
console_handler.setFormatter(formatter)
# set logger
logger = logging.getLogger('logger')
logger.setLevel('DEBUG') # 设置了这个才会把debug以上的输出到控制台
logger.addHandler(file_handler)
logger.addHandler(console_handler)
self.logger = logger
def calc_angles(arr):
beta = np.arctan(-arr[2, 0] / (np.sqrt(arr[0, 0] * arr[0, 0] + arr[1, 0] * arr[1, 0]))) / pi * 180
alpha = np.arctan(arr[1, 0] / np.cos(beta) / (arr[0, 0] / np.cos(beta))) / pi * 180
if alpha < 0:
alpha += 180
gamma = np.arctan(arr[2, 1] / np.cos(beta) / (arr[2, 2] / np.cos(beta))) / pi * 180
return [beta, alpha, gamma]
def probe_angle(ee_position):
arr1 = np.asarray(ee_position).reshape((4, 4)).T
arr2 = np.array([[-1e-9, -1, -1e-9],
[-1, 1e-9, -1e-9],
[1e-9, 1e-9, -1]])
[beta1, alpha1, gamma1] = calc_angles(arr1)
[beta2, alpha2, gamma2] = calc_angles(arr2)
angle_diffs = np.array([gamma1 - gamma2, beta1 - beta2, alpha1 - alpha2])
return [np.around(angle_diffs, 1), [beta1, alpha1, gamma1]]
def calib_angle(ee_position):
T1 = np.asarray(ee_position).reshape((4, 4)).T
# angle = np.array([0.05675555627186434, 0.7781619195601319, 0.625494071737504, 0.0,
# 0.0795119019880618, -0.628036939167217, 0.7741107546622216, 0.0,
# 0.9952168930813586, 0.00579913682921899, -0.09751772012062157, 0.0,
# 0.29798168921190665, -0.1448472095247668, 0.7506857661469335, 1.0])
# JN
# angle = np.array([0.04708377235658972, 0.7813972391244601, 0.6222551510990844, 0.0,
# 0.07447851970792423, -0.6239582615150934, 0.7779004036437948, 0.0,
# 0.9961104700196606, 0.009718157013226774, -0.08757561841903876, 0.0,
# 0.2980300396445975, -0.12727154607479305, 0.7676168006362032, 1.0])
# # LSQ
# angle = np.array([ 0.14942991, 0.79438478, 0.58874742, 0,
# 0.1389825 , -0.60639603, 0.78292255, 0,
# 0.97895585, -0.03516646, -0.20101929, 0,
# 0.35941167, -0.25011404, 0.80949064, 1])
# # SZG
# angle = np.array([-0.09915866, 0.82699498, 0.55339576, 0,
# 0.10848553, -0.54383705, 0.83214912, 0,
# 0.98914026, 0.14255023, -0.03579083, 0,
# 0.31304733, -0.12770589, 0.81040914, 1])
# SY
angle = np.array([ 0.30909544, 0.77366372, 0.5530863 , 0. ,
0.18894045, -0.61992756, 0.76157162, 0. ,
0.93207377, -0.13089794, -0.33779316, 0. ,
0.35927308, -0.21256274, 0.82335111, 1. ])
T2 = angle.reshape((4, 4)).T
T1_inv = np.linalg.inv(T1) # 计算逆矩阵
T_diff = np.dot(T1_inv, T2) # 计算差值矩阵
R_diff = T_diff[:3, :3] # 提取旋转矩阵和平移向量
r = Rotation.from_matrix(R_diff) # 计算旋转矩阵的欧拉角表示
rpy_diff = r.as_euler('xyz', degrees=False)
res = [0, 0, 0] + list(rpy_diff)
print(f'** res : {res} **')
return res
# --------------------------------- lsq 0621 -------------------------------------
def transformedMatrix(originalMatrix):
rotation_z = np.eye(4)
rotation_z[:3, :3] = rotate_3d(np.pi, 'z')
rotation_y = np.eye(4)
rotation_y[:3, :3] = rotate_3d(-np.pi / 2, 'y')
return np.dot(np.dot(originalMatrix, rotation_z), rotation_y)
def rotate_3d(theta, axis):
if axis == 'x':
rotation_matrix = np.array([[1, 0, 0],
[0, cos(theta), -sin(theta)],
[0, sin(theta), cos(theta)]])
elif axis == 'y':
rotation_matrix = np.array([[cos(theta), 0, sin(theta)],
[0, 1, 0],
[-sin(theta), 0, cos(theta)]])
elif axis == 'z':
rotation_matrix = np.array([[cos(theta), -sin(theta), 0],
[sin(theta), cos(theta), 0],
[0, 0, 1]])
else:
raise ValueError("Invalid axis value. Please choose 'x', 'y', or 'z'.")
return rotation_matrix
# ---------------------------------
def main():
# params
script_code = 'CV' # identify vision from RL or other fields
ros_rate = 30 # Hz, for ros
max_frame = 30
camera_id = 3 # for StreamVideo
num_classes = 6
root = '/media/robotics1/WD_G_2T'
# init object
aisono = ArtificialSonographer(camera_id) # AI Sonographer
aisono.init_log(log_root=join(root, 'logs'))
aisono.logger.info(f'\n{"=" * 64}\n {"Auto Echocardiography".center(60)}\n{"-" * 64}\n\n'
'Execute keyboard commands in "discrete_keyboard" terminal\n'
'1. To visualise video stream, press "c" \n'
'2. To initiate auto scanning, press "s" \n'
'3. To calibrate probe pose, press "a" \n'
'4. To bridge the gap between probe and chest, press "b"')
model_path = '/media/robotics1/WD_2T/sunyu_data/models/6.29.pth'
aisono.logger.info(f'\n {"-- >>":>21} {"preparing models ... ":^20}{"<< --":<21} ')
aisono.model = aisono.__load_model__(model_path, num_classes)
aisono.logger.info(f' {"-- >>":>21} {"model ready ":^20}{"<< --":<21}\n')
date = time.strftime('%Y%m%d-%H%M', time.localtime())
save_root = join(root, 'inference_record', f'{script_code}_{date}')
# ros
rospy.init_node('inference_vision', anonymous=True)
aisono.pub = rospy.Publisher('keyboard', Int64, queue_size=10)
aisono.pub2 = rospy.Publisher('vector_topic', Float64MultiArray, queue_size=10)
rospy.Subscriber("keyboard", Int64, aisono.keyboard)
rospy.Subscriber("/franka_state_controller/franka_states",
FrankaState, aisono.Franka_cb)
# rospy.Subscriber('/my_heart_joint_velocity_setpoint_controller_vision/action_finished',
# Int64, aisono.action_control)
rospy.Subscriber('/my_heart_cartesian_impedance_controller/action_finished',
Int64, aisono.action_control)
rate = rospy.Rate(ros_rate)
# counters
input_tensor = torch.zeros((max_frame, 3, 224, 224))
frame_count = 0
angle_count = 0
img_list = []
aisono.logger.info(f'\n{" | ":->24} {"Stage":>9} {aisono.stage_ind:<4} {" | ":-<24}')
while not rospy.is_shutdown():
if aisono.flag_autoinf:
aisono.flag_infrn = True
# section 1: video stream
frame, angles = aisono.streaming(aisono.flag_position, aisono.ee_position)
frame_show = copy.deepcopy(frame)
if aisono.flag_stream:
cv2.putText(frame_show, f'{angles[0]}, {angles[1]}, {angles[2]}', (10, 50),
cv2.FONT_HERSHEY_SIMPLEX, 0.75, (173, 206, 230), 1, 8, 0)
cv2.imshow("UltrasoundStream", frame_show) # 173, 206, 230; 205, 179, 139
cv2.waitKey(1)
else:
cv2.destroyAllWindows()
cv2.waitKey(1)
# section 1.1: calib angle
if aisono.flag_angle and not aisono.flag_move and not aisono.flag_moving:
current_angles = calib_angle(aisono.ee_position)
vector_msg = Float64MultiArray(data=current_angles)
print(f'=> publish message: {current_angles}')
aisono.pub2.publish(vector_msg)
aisono.flag_angle = False
aisono.logger.info(f'{" ":<10}{"-" * 42}{" ":>10}\n'
f'{" ":<10}* Current state of [Angle Switch]: '
f'{aisono.flag_angle:^4} *{" ":>10}\n{" ":<10}{"-" * 42}{" ":>10}')
# section 2: inference (every 30 frames)
if not aisono.flag_infrn:
continue
if aisono.flag_scan:
if frame_count < max_frame:
img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
img_list.append(img)
input_tensor[frame_count, :, :, :] = aisono.trans(img)
frame_count += 1
continue
elif frame_count == max_frame: # 30 frames gathered, ready for publishing action
frame_count = 0
aisono.action_count += 1
aisono.flag_scan = False
aisono.flag_vote = True
else:
aisono.logger.info(f'** ERROR: abnomal frame count: {frame_count} **')
# section 3: max vote, publish stride and action
if aisono.flag_vote:
aisono.flag_vote = False
input_mean = input_tensor.mean(dim=0)
fuse_tensor = torch.zeros((max_frame, 3, 224, 224))
for i in range(input_tensor.shape[0]):
fuse_tensor[i, :, :, :] = input_mean * 0.6 + input_tensor[i, :] * 0.4
with torch.no_grad():
outputs = aisono.model(fuse_tensor.to(aisono.device))
# 去掉极值，取平均
outputs, _ = torch.sort(outputs, dim=0)
outputs = outputs[5:25, :]
outputs = torch.mean(outputs, dim=0).tolist()
# outputs = [round(x, 4) for x in outputs]
# 转换格式 （m to mm, degree to arc）
outputs = [x / 1000 for x in outputs[:3]] + [0, 0, 0]
outputs = [round(x, 5) for x in outputs]
stop_inference_signal = [x for x in outputs[:3] if abs(x) > 0.002]
# print(stop_inference_signal)
if len(stop_inference_signal) == 0:
aisono.stop_signal += 1
if aisono.stop_signal == 2:
aisono.flag_autoinf = False
aisono.flag_infrn = False
vector_msg = Float64MultiArray(data=outputs)
aisono.logger.info(f'=> publish message: {outputs}')
aisono.pub2.publish(vector_msg)
# save images with action
save_dir = join(save_root, f'{aisono.action_count:03d}-{chr(aisono.action)}')
os.makedirs(save_dir, exist_ok=True)
for img_ind, img in enumerate(img_list):
img.save(join(save_dir, f'{img_ind:03d}.jpg'))
img_list = []
input_tensor = torch.zeros((max_frame, 3, 224, 224))
aisono.flag_scan = True
aisono.flag_infrn = False
rate.sleep()
aisono.cap.release()
aisono.logger.info(f'{"-" * 64}\n {"Auto Echocardiography Complete":^60}\n{"=" * 64} ')
class Infer():
def __init__(self, model_path, num_class):
self.model_path = model_path
self.num_class = num_class
self.device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
def __load_model__(self):
self.model = models.resnet34()
self.model.fc = nn.Sequential(
nn.Linear(model.fc.in_features, 256),
nn.Dropout(0.4),
nn.Linear(256, self.num_class))
self.model.load_state_dict(torch.load(self.model_path))
self.model.to(self.device)
self.model.eval()
return self.model
def __call__(self, tensor):
model = self.__load_model__()
tensor = tensor.cuda() if torch.cuda.is_available() else tensor
outputs = model(tensor)
outputs, _ = torch.sort(outputs, dim=0)
outputs = outputs[5:25, :]
outputs = torch.mean(outputs, dim=0).tolist()
outputs = [x / 1000 for x in outputs[:3]] + [0, 0, 0]
return outputs