-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_split1_org.py
223 lines (202 loc) · 7.82 KB
/
train_split1_org.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
import numpy as np
import time, sys, os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
# 0 = all messages are logged(default behavior)
# 1 = INFO messages are not printed
# 2 = INFO and WARNING messages arenot printed
# 3 = INFO, WARNING, and ERROR messages arenot printed
import src.dataset_loader.KITTI_dataset_p2_mat as kitti
import src.net_core.darknet as Darknet
import src.module.nolbo as nolbo
import tensorflow as tf
import cv2
tf.get_logger().warning('test')
# WARNING:tensorflow:test
tf.get_logger().setLevel('ERROR')
tf.get_logger().warning('test')
imageSizeAndBatchListKITTI = [
# # [384,128,64],
# # [480,160,64],
# [576,192,64,26],
# [672,224,64,24],
# [768,256,64,22],
# [864,288,52,14],
# [960,288,50,14],
# [960,320,48,14],
# [960,352,46,12],
# [1056,352,46,10],
# [1088,352,46,10],
# [1088,384,44,10],
# [1152,384,42,8],
[1216,448,42,6],
# [1216,448,64,8],
# [1248,384,42,8],
# [1248,448,42,8],
# # #############################
# [160,128,64],
# [288,288,56],
# [320,256,48],
# [320,320,48],
# [416,416,40],
# [480,384,40],
# [448,448,40],
# [640,512,32],
# [800,640,24],
]
def train(
training_epoch = 1000,
learning_rate = 1e-4, solver='adam',
config = None, predictor_num=5,
save_path = None, load_path = None,
load_encoder_backbone_path = None, load_encoder_backbone_name = None,
load_decoder_path = None, load_decoder_name = None,
data_path='',
):
# data_path = '/home/yonsei/pyws/NOLBO_grid_anchor_no3Ddec/NOLBO_2D_3D_IoU_exp_viewing/data/kitti_split1_org'
data_loader = kitti.dataLoader(predNumPerGrid=predictor_num,
KITTIImagePath=os.path.join(data_path, 'training/image_2/'),
KITTIAnnotPath=os.path.join(data_path, 'training/3dv_2/'),
KITTI_calib_path=os.path.join(data_path, 'training/calib/'),
KITTI3DShapePath='/media/yonsei/4TB_HDD/dataset/PASCAL3D+_release1.1/CAD/car/',
KITTI_anchor_path=data_path
)
model = nolbo.nolbo(nolbo_structure=config, solver=solver,
learning_rate=learning_rate,
IoU2D_loss=False, IoU3D_loss=False)
if load_path != None:
print('load weights...')
model.loadModel(load_path=load_path)
# model.loadEncoder(load_path=load_path)
# model.loadEncoderBackbone(load_path=load_path)
# model.loadEncoderHead(load_path=load_path)
# model.loadEncoderHeadTop(load_path=load_path)
# model.loadDecoder(load_path=load_path)
# model.loadPriornet(load_path=load_path)
print('done!')
if load_encoder_backbone_path != None:
print('load encoder backbone weights...')
model.loadEncoderBackbone(
load_path=load_encoder_backbone_path,
file_name=load_encoder_backbone_name
)
print('done!')
if load_decoder_path != None:
print('load decoder weights...')
model.loadDecoder(
load_path=load_decoder_path,
file_name=load_decoder_name
)
print('done!')
loss = np.zeros(14)
epoch, iteration, run_time = 0., 0., 0.
item_num = 0
print('start training...')
while epoch < training_epoch:
start_time = time.time()
periodOfImageSize = 3
if int(iteration) % (periodOfImageSize * len(imageSizeAndBatchListKITTI)) == 0:
np.random.shuffle(imageSizeAndBatchListKITTI)
image_col, image_row, _, batch_size = imageSizeAndBatchListKITTI[
int(iteration) % int((periodOfImageSize * len(imageSizeAndBatchListKITTI)) / periodOfImageSize)]
image_size = image_col, image_row
batch_data = data_loader.getNextBatch(batchSize=batch_size,
imageSize=image_size,
gridSize=(int(image_size[0] / 32), int(image_size[1] / 32)),
augmentation=True, dist_type='s')
# images = batch_data[2]
# for image in images:
# cv2.imwrite(os.path.join('./test/image/', '{:03d}.png'.format(item_num)), image)
# item_num += 1
epoch_curr = data_loader.epoch
data_start = data_loader.dataStart
data_length = data_loader.dataLength
if epoch_curr != epoch:
print('')
iteration = 0
loss = loss * 0.
run_time = 0.
if save_path != None:
print('save model...')
model.saveModel(save_path=save_path)
epoch = epoch_curr
# try:
loss_temp = model.fit(inputs=batch_data)
end_time = time.time()
loss = (loss * iteration + np.array(loss_temp)) / (iteration + 1.0)
run_time = (run_time * iteration + (end_time - start_time)) / (iteration + 1.0)
sys.stdout.write(
"Ep:{:03d} it:{:04d} rt:{:.2f} ".format(int(epoch + 1), int(iteration + 1), run_time))
sys.stdout.write("cur/tot:{:04d}/{:04d} b:{:02d} ".format(data_start, data_length, len(batch_data[0])))
sys.stdout.write(
"obj:{:.2f}, noobj:{:.2f}, b2D:{:.3f}, b3DI:{:.3f}, ".format(loss[0], loss[1], loss[2], loss[3]))
sys.stdout.write(
"b3Dd:{:.3f}, loc:{:.3f} ".format(loss[4], loss[5]))
sys.stdout.write(
"sc:{:.2f}, sc1:{:.2f} ".format(loss[6], loss[7]))
sys.stdout.write(
"ca:{:.2f}, sh:{:.2f} ".format(loss[8], loss[9]))
sys.stdout.write(
"op:{:.2f}, np:{:.2f} ".format(loss[10], loss[11]))
sys.stdout.write(
"pr:{:.3f}, rc:{:.3f} \r".format(loss[12], loss[13]))
sys.stdout.flush()
if np.sum(loss) != np.sum(loss):
print('')
print('NaN')
return
iteration += 1.0
# except:
# # pass
# print('save model...')
# model.saveModel(save_path=save_path)
# print(image_col, image_row, len(batch_data[0]), len(batch_data[-4]))
# return
predictor_num = 12
latent_dim = 64
category_num = kitti.category_num
car_instance_num = kitti.car_instance_num
config = {
'category_num':category_num,
'encoder_backbone':{
'name' : 'nolbo_backbone',
'predictor_num': predictor_num,
'bbox2DXY_dim':2, 'bbox3D_dim':3, 'localXYZ_dim':1, 'orientation_dim':1,
'latent_dim': latent_dim,
'activation' : 'lrelu',
},
'encoder_head':{
'name' : 'nolbo_head',
'output_dim' : predictor_num*(1+2+3+1+2+ 2*latent_dim), # objness + b2Dxy + lhw + z + sincos
'activation':'relu',
},
'decoder':{
'name':'decoder',
'input_dim' : latent_dim,
'output_shape':[64,64,64,1],
'filter_num_list':[512,256,128,64,1],
'filter_size_list':[4,4,4,4,4],
'strides_list':[1,2,2,2,2],
'activation':'relu',
'final_activation':'sigmoid'
},
'prior' : {
'name' : 'priornet',
'input_dim' : car_instance_num, # class num (one-hot vector)
'unit_num_list' : [latent_dim//2, latent_dim],
'core_activation' : 'relu',
'const_log_var' : 0.0,
}
}
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
if __name__ == '__main__':
sys.exit(train(
training_epoch=20, learning_rate=1e-4, solver='adam',
config=config, predictor_num=predictor_num,
save_path='./weights/kitti_split1_org/',
load_path='./weights/kitti_split1_org/',
# load_encoder_backbone_path='./weights/yolov2/',
# load_encoder_backbone_name='nolbo_backbone',
# load_decoder_path='./weights/AE3D/',
# load_decoder_name='decoder3D',
data_path='./data/kitti_split1',
))