-
Notifications
You must be signed in to change notification settings - Fork 2
/
re-id.py
244 lines (207 loc) · 8.89 KB
/
re-id.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
import sys
sys.path.insert(0, "mxnet/python/")
import mxnet as mx
import logging
import numpy as np
import argparse
import time
import random
from mxnet.optimizer import SGD
import loss_layers
from verifi_iterator import verifi_iterator
import importlib
def build_network(symbol, num_id, num_model):
'''
network structure
'''
# concat = internals["ch_concat_5b_chconcat_output"]
pooling = mx.symbol.Pooling(
data=symbol, kernel=(1, 1), global_pool=True,
pool_type='avg', name='global_pool')
flatten = mx.symbol.Flatten(data=pooling, name='flatten')
fc1 = mx.symbol.FullyConnected(
data=flatten, num_hidden=num_id, name='cls_fc1')
softmax1 = mx.symbol.SoftmaxOutput(data=fc1, name='softmax1')
fc2 = mx.symbol.FullyConnected(
data=flatten, num_hidden=num_model, name='cls_fc2')
softmax2 = mx.symbol.SoftmaxOutput(data=fc2, name='softmax2')
l2 = mx.symbol.L2Normalization(data=flatten, name='l2_norm')
verifi = mx.symbol.Custom(
data=l2, grad_scale=1.0, threshd=args.verifi_threshd,
op_type='verifiLoss', name='verifi')
triplet = mx.symbol.Custom(
data=l2, grad_scale=1.0, threshd=args.triplet_threshd,
op_type='tripletLoss', name='triplet')
outputs = [softmax1, softmax2]
if args.verifi:
outputs.append(verifi)
if args.triplet:
outputs.append(triplet)
return mx.symbol.Group(outputs)
class Multi_Metric(mx.metric.EvalMetric):
"""Calculate accuracies of multi label"""
def __init__(self, num=None, cls=1):
super(Multi_Metric, self).__init__('multi-metric', num)
self.cls = cls
def update(self, labels, preds):
# mx.metric.check_label_shapes(labels, preds)
# if self.num != None:
# assert len(labels) == self.num
for i in range(self.cls):
pred_label = mx.nd.argmax_channel(
preds[i]).asnumpy().astype('int32')
label = labels[i].asnumpy().astype('int32')
mx.metric.check_label_shapes(label, pred_label)
if self.num is None:
self.sum_metric += (pred_label.flat == label.flat).sum()
self.num_inst += len(pred_label.flat)
else:
self.sum_metric[i] += (pred_label.flat == label.flat).sum()
self.num_inst[i] += len(pred_label.flat)
for i in range(self.cls, len(preds)):
pred = preds[i].asnumpy()
if self.num is None:
self.sum_metric += np.sum(pred)
self.num_inst += len(pred)
else:
self.sum_metric[i] += np.sum(pred)
self.num_inst[i] += len(pred)
def get_imRecordIter(name, input_shape, batch_size, kv, shuffle=False, aug=False):
dataiter = mx.io.ImageRecordIter(
path_imglist="%s/%s.lst" % (args.data_dir, name),
path_imgrec="%s/%s.rec" % (args.data_dir, name),
mean_img="models/vid_mean.bin",
rand_crop=aug,
rand_mirror=aug,
prefetch_buffer=4,
preprocess_threads=3,
shuffle=shuffle,
label_width=2,
data_shape=input_shape,
batch_size=batch_size / 2,
num_parts=kv.num_workers,
part_index=kv.rank)
return dataiter
def get_iterators(batch_size, input_shape, train, test, kv, gpus=1):
'''
use image list and rec file to generate data iterators
'''
train_dataiter1 = get_imRecordIter(
'%s-even' % train, input_shape, batch_size,
kv, shuffle=False, aug=True)
train_dataiter2 = get_imRecordIter(
'%s-rand' % train, input_shape, batch_size,
kv, shuffle=True, aug=True)
val_dataiter1 = get_imRecordIter(
'%s-even' % test, input_shape, batch_size,
kv, shuffle=False, aug=False)
val_dataiter2 = get_imRecordIter(
'%s-rand' % test, input_shape, batch_size,
kv, shuffle=False, aug=False)
return verifi_iterator(
train_dataiter1, train_dataiter2, use_verifi=args.verifi, gpus=gpus), \
verifi_iterator(
val_dataiter1, val_dataiter2, use_verifi=args.verifi, gpus=gpus)
# return train_dataiter2, val_dataiter2
def parse_args():
parser = argparse.ArgumentParser(
description='single domain car recog training')
parser.add_argument('--gpus', type=str, default='0',
help='the gpus will be used, e.g "0,1,2,3"')
parser.add_argument('--data-dir', type=str,
default="/mnt/hdd/lytton/mx_data/VehicleID",
help='data directory')
parser.add_argument('--num-examples', type=int, default=119698,
help='the number of training examples')
parser.add_argument('--num-id', type=int, default=13164,
help='the number of training examples')
parser.add_argument('--batch-size', type=int, default=32,
help='the batch size')
parser.add_argument('--lr', type=float, default=.01,
help='the initial learning rate')
parser.add_argument('--num-epoches', type=int, default=100,
help='the number of training epochs')
parser.add_argument('--mode', type=str, default='tmp',
help='save names of model and log')
parser.add_argument('--verifi-label', action='store_true', default=False,
help='if add verifi label')
parser.add_argument('--verifi', action='store_true', default=False,
help='if use verifi loss')
parser.add_argument('--triplet', action='store_true', default=False,
help='if use triplet loss')
parser.add_argument('--verifi-threshd', type=float, default=0.9,
help='verification threshold')
parser.add_argument('--triplet-threshd', type=float, default=0.9,
help='triplet threshold')
parser.add_argument('--train-file', type=str, default="train-split1",
help='train file')
parser.add_argument('--test-file', type=str, default="train-split2",
help='test file')
parser.add_argument('--kv-store', type=str,
default='device', help='the kvstore type')
parser.add_argument('--network', type=str,
default='inception-bn', help='network name')
parser.add_argument('--model-load-epoch', type=int, default=126,
help='load the model on an epoch using the model-load-prefix')
parser.add_argument('--model-load-prefix', type=str, default="inception-bn",
help='load model prefix')
return parser.parse_args()
def load_checkpoint(prefix, epoch):
# ssymbol = sym.load('%s-symbol.json' % prefix)
save_dict = mx.nd.load('%s-%04d.params' % (prefix, epoch))
arg_params = {}
aux_params = {}
for k, v in save_dict.items():
tp, name = k.split(':', 1)
if tp == 'arg':
arg_params[name] = v
if tp == 'aux':
aux_params[name] = v
return (arg_params, aux_params)
args = parse_args()
print args
batch_size = args.batch_size
num_epoch = args.num_epoches
devices = [mx.gpu(int(i)) for i in args.gpus.split(',')]
lr = args.lr
num_images = args.num_examples
arg_params, aux_params = load_checkpoint(
'models/%s' % args.model_load_prefix, args.model_load_epoch)
symbol = importlib.import_module(
'symbol_' + args.network).get_symbol()
net = build_network(symbol, num_id=args.num_id, num_model=251)
kv = mx.kvstore.create(args.kv_store)
train, val = get_iterators(
batch_size=batch_size, input_shape=(3, 224, 224),
train=args.train_file, test=args.test_file, kv=kv, gpus=len(devices))
stepPerEpoch = int(num_images * 2 / batch_size)
lr_scheduler = mx.lr_scheduler.MultiFactorScheduler(
step=[stepPerEpoch * x for x in [50, 75]], factor=0.1)
init = mx.initializer.Xavier(
rnd_type='gaussian', factor_type='in', magnitude=2)
arg_names = net.list_arguments()
sgd = SGD(learning_rate=args.lr, momentum=0.9,
wd=0.0001, clip_gradient=10, lr_scheduler=lr_scheduler,
rescale_grad=1.0 / batch_size)
# args_lrscale = {}
# index = 0
# for name in arg_names:
# if name != 'data' and name != 'softmax_label':
# args_lrscale[index] = 1.0 if name.startswith('car_fc') else 1.0
# index += 1
# sgd.set_lr_mult(args_lrscale)
logging.basicConfig(filename='log/%s.log' % args.mode, level=logging.DEBUG)
logger = logging.getLogger()
logger.setLevel(logging.DEBUG)
logging.info(args)
model = mx.model.FeedForward(
symbol=net, ctx=devices, num_epoch=num_epoch, arg_params=arg_params,
aux_params=aux_params, initializer=init, optimizer=sgd)
prefix = 'models/%s' % args.mode
num = 2
if args.verifi:
num += 1
if args.triplet:
num += 1
model.fit(X=train, eval_data=val, eval_metric=Multi_Metric(num=num, cls=2), logger=logger, epoch_end_callback=mx.callback.do_checkpoint(prefix),
batch_end_callback=mx.callback.Speedometer(batch_size=batch_size))