-
Notifications
You must be signed in to change notification settings - Fork 8
/
train_chunk_imshuffle.py
170 lines (139 loc) · 5.91 KB
/
train_chunk_imshuffle.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
import h5py
import pickle
import numpy as np
import tempfile
import time
import os
os.environ['GLOG_minloglevel'] = '2'
import caffe
caffe.set_mode_gpu()
def static_vars(**kwargs):
def decorate(func):
for k in kwargs:
setattr(func, k, kwargs[k])
return func
return decorate
def create_fg_net():
f = tempfile.NamedTemporaryFile(mode='w+', delete=False)
f.write(
"""
name: 'nn'
layer { name: "data" type: "DummyData" top: "data" dummy_data_param { shape { dim: 1 dim: 1 dim: 1 dim: 1024 } } }
layer { name: "label" type: "DummyData" top: "label" dummy_data_param { shape { dim: 1 dim: 1 dim: 1 dim: 1 } } }
layer { name: 'ip' type: 'InnerProduct' bottom: 'data' top: 'ip1' inner_product_param { num_output: 1000 weight_filler {type: "xavier"} } }
layer { name: 'drop' type: 'Dropout' bottom: 'ip1' top: 'ip1' dropout_param { dropout_ratio: 0.5 } }
layer { name: 'ip2' type: 'InnerProduct' bottom: 'ip1' top: 'ip2' inner_product_param { num_output: 201 weight_filler {type: "xavier"} } }
layer { name: 'loss' type: 'SoftmaxWithLoss' bottom: 'ip2' bottom: 'label' top: 'loss' loss_param { normalize: true ignore_label: 0} }
""")
f.close()
return f.name
def create_solver(netf):
f = tempfile.NamedTemporaryFile(mode='w+', delete=False)
f.write("""
net: '""" + netf + """'
base_lr: 0.01
momentum: 0.9
weight_decay: 0.00005
lr_policy: 'fixed'
display: 200000
snapshot: 1000
max_iter: 200000
snapshot_prefix: "chunk_ims"
""")
f.close()
return f.name
def create_deploy():
f = tempfile.NamedTemporaryFile(mode='w+', delete=False)
f.write(
"""
name: 'nn'
layer { name: "data" type: "DummyData" top: "data" dummy_data_param { shape { dim: 1 dim: 1 dim: 1 dim: 1024 } } }
layer { name: 'ip' type: 'InnerProduct' bottom: 'data' top: 'ip1' inner_product_param { num_output: 1000 } }
layer { name: 'ip2' type: 'InnerProduct' bottom: 'ip1' top: 'ip2' inner_product_param { num_output: 201 } }
layer { name: 'loss' type: 'Softmax' bottom: 'ip2' top: 'prob' }
""")
f.close()
return f.name
def classify_video(model, subset='validation', ANNOTATION = 'actNet200-V1-3.pkl', FEATURE = '/home/EXTRA/DATASETS/actnet/shuffle_imagenet/'):
nf = create_deploy()
net = caffe.Net(nf, model, caffe.TEST)
with open(ANNOTATION, 'rb') as f:
gt = pickle.load(f)['database']
with h5py.File(FEATURE + 'ImageNetShuffle2016_features.h5', 'r') as f:
mbh = np.asarray(f['features'])
with open(FEATURE + 'ImageNetShuffle2016_quids.txt', 'r') as f:
lines = f.readlines()
idmap = {}
for line in lines:
temp = line.split(',')
idmap[temp[2].split('.')[0]] = int(temp[3].replace(')\n',''))
scores = {}
for vid, vitem in gt.iteritems():
if vitem['subset']==subset:
ind = idmap['v_'+vid]
feat = mbh[ind, :]
net.blobs['data'].data[...] = feat
net.forward()
prob = net.blobs['prob'].data
scores[vid] = np.copy(prob)
return scores
if __name__ == '__main__':
import logging
logging.basicConfig(format='%(asctime)s %(message)s', filename='train_chunk_imshuffle'+time.strftime('_%Y_%m_%d.log'), level=logging.DEBUG)
logging.getLogger()
logging.info(' 201 classifier with imagenetshuffle')
FEATURE = os.environ['ACTNET_HOME'] + '/shuffle_imagenet/'
with open('actNet200-V1-3.pkl', 'rb') as f:
gt = pickle.load(f)['database']
with h5py.File(FEATURE + 'ImageNetShuffle2016_features.h5', 'r') as f:
mbh = np.asarray(f['features'])
with open(FEATURE + 'ImageNetShuffle2016_quids.txt', 'r') as f:
lines = f.readlines()
idmap = {}
for line in lines:
temp = line.split(',')
idmap[temp[2].split('.')[0]] = temp[3].replace(')\n','')
indexes = {'training':[],'validation':[],'testing':[]}
labels = {'training':[],'validation':[]}
for vid, vitem in gt.iteritems():
if vitem['subset']!='testing':
labels[vitem['subset']].append(vitem['annotations'][0]['class'])
indexes[vitem['subset']].append(idmap['v_'+vid])
############# Train and test a nn #############
nf = create_fg_net()
sf = create_solver(nf)
solver = caffe.get_solver(sf)
logging.debug('')
max_ep = 20
max_iter = 50
batch = 200
# test_feat = mbh[np.array(indexes['testing'], dtype=int),:]
list = indexes['training']
for ep in range(max_ep):
logging.info('Epoch %d ' % (ep + 1))
loss = 0
for it in range(max_iter):
############# Train #############
x = mbh[np.array(indexes['training'][it*batch:(it+1)*batch], dtype=int),:]
y = np.array(labels['training'][it*batch:(it+1)*batch], dtype=int)
x = x[:, np.newaxis, np.newaxis, :]
y = y[:, np.newaxis, np.newaxis, np.newaxis]
solver.net.blobs['data'].reshape(*x.shape)
solver.net.blobs['data'].data[...] = x
solver.net.blobs['label'].reshape(*y.shape)
solver.net.blobs['label'].data[...] = y
solver.step(1)
loss += solver.net.blobs['loss'].data
logging.info('Train Loss: %f '%(loss/max_iter))
############# Test #############
val_feat = mbh[np.array(indexes['validation'], dtype=int), :]
val_lb = np.array(labels['validation'])
val_feat = val_feat[:, np.newaxis, np.newaxis, :]
val_lb = val_lb[:, np.newaxis, np.newaxis, np.newaxis]
solver.net.blobs['data'].reshape(*val_feat.shape)
solver.net.blobs['data'].data[...] = val_feat
solver.net.blobs['label'].reshape(*val_lb.shape)
solver.net.forward()
prob = solver.net.blobs['ip2'].data
pred = prob.argmax(axis=1)
logging.info('Test on validation set ACC %f'%(np.sum(pred.squeeze()==val_lb.squeeze())*1.0/val_lb.shape[0]))