-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutil_init.py
335 lines (284 loc) · 18.6 KB
/
util_init.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
# This file is for Init. Experiment （5 seen + 5 unseen）
import numpy as np
import scipy.io as sio
import torch
from sklearn import preprocessing
def weights_init(m):
classname = m.__class__.__name__
if classname.find('Linear') != -1:
m.weight.data.normal_(0.0, 0.02)
if m.bias is not None:
m.bias.data.fill_(0)
elif classname.find('BatchNorm') != -1:
m.weight.data.normal_(1.0, 0.02)
m.bias.data.fill_(0)
def map_label(label, classes):
mapped_label = torch.LongTensor(label.size())
for i in range(classes.size(0)):
mapped_label[label == classes[i]] = i
return mapped_label
class DATA_LOADER(object):
def __init__(self, opt):
self.read_matdataset(opt)
self.index_in_epoch = 0
self.epochs_completed = 0
def read_matdataset(self, opt):
if opt.dataset == "ucf101":
# load visual features for ucf101
matcontent = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + opt.dataset
+ "_" + opt.action_embedding + ".mat")
feature = matcontent['features'].T
label = matcontent['labels'].astype(int).squeeze() - 1
# load action dataset splits and semantics
# for inistal exp. (20 classes)
#matcontent = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + "att_split_6classes.mat")
matcontent = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + "att_split_20classes.mat")
# trainval_loc = matcontent['trainval_loc'].squeeze() - 1
train_loc = matcontent['train_loc'].squeeze() - 1
# val_unseen_loc = matcontent['val_loc'].squeeze() - 1
#test_seen_loc = matcontent['test_seen_loc'].squeeze() - 1
test_unseen_loc = matcontent['test_unseen_loc'].squeeze() - 1
if opt.class_embedding == "att":
self.attribute = torch.from_numpy(matcontent['original_att'].T).float()
self.attribute /= self.attribute.pow(2).sum(1).sqrt().unsqueeze(1).expand(self.attribute.size(0),
self.attribute.size(1))
elif opt.class_embedding == "wv":
if opt.object:
# att_all: 300 + 900d
print("with object semantics:")
#print("append 3 objects - 1200d")
self.attribute = torch.from_numpy(matcontent['att_all'].T).float() # (101, 1200)
# action vector
#self.action = torch.vstack((self.attribute[:10, :300],
# self.attribute[10, :300],
# self.attribute[11:13, :300],
# self.attribute[13, :300],
# self.attribute[14:32, :300],
# self.attribute[32:34, :300],
# self.attribute[34:89, :300],
# self.attribute[89, :300],
# self.attribute[90:, :300]))
self.action = self.attribute[:, :300]
# best object vector
self.best_obj = torch.vstack((self.attribute[:10, 300:600], # 0-9 (1st)
self.attribute[10, 600:900], # 10 (2nd)
self.attribute[11:13, 300:600], # 11, 12 (1st)
self.attribute[13, 600:900], # 13 (2nd)
self.attribute[14:32, 300:600], # 14-31 (1st)
self.attribute[32:34, 900:], # 32, 33 (3rd)
self.attribute[34:89, 300:600], # 34-88 (1st)
self.attribute[89, 900:], # 89 (3rd)
self.attribute[90:, 300:600])) # 90-100 (1st)
# The first object.
self.obj1 = self.attribute[:, :300]
# The second object.
self.obj2 = self.attribute[:, 600:900]
# The third object.
self.obj3 = self.attribute[:, 900:]
# Different cases:
###############################################################################################
# Case: Best object
# seen: action + 1st object; Unseen: action + best
# seen: 3,8,20,26,55.78.88.95.98.100
# unseen (best object): 11 (2), 13(1), 14(2), 33(3), 34(3), 42(1), 73(1), 85(1), 90(3), 92(1)
#print("Append best object.")
#self.attribute = torch.vstack((self.attribute[:10, :600], # 0-9 (1st)
# torch.hstack((self.attribute[10, :300], self.attribute[10, 600:900])), # 10 (2nd)
# self.attribute[11:13, :600], # 11, 12 (1st)
# torch.hstack((self.attribute[13, :300], self.attribute[13, 600:900])), # 13 (2nd)
# self.attribute[14:32, :600], # 14-31 (1st)
# torch.hstack((self.attribute[32:34, :300], self.attribute[32:34, 900:])), # 32, 33 (3rd)
# self.attribute[34:89, :600], # 34-88 (1st)
# torch.hstack((self.attribute[89, :300], self.attribute[89, 900:])), # 89 (3rd)
# self.attribute[90:, :600]) # 90-100 (1st)
# )
# Case: Repalce with best object
# seen: 1st object; Unseen: best object
# seen: 3,8,20,26,55.78.88.95.98.100
# unseen (best object): 11 (2), 13(1), 14(2), 33(3), 34(3), 42(1), 73(1), 85(1), 90(3), 92(1)
#print("Replace action with best object.")
#self.attribute = torch.vstack((self.attribute[:10, 300:600], # 0-9 (1st)
# self.attribute[10, 600:900], # 10 (2nd)
# self.attribute[11:13, 300:600], # 11, 12 (1st)
# self.attribute[13, 600:900], # 13 (2nd)
# self.attribute[14:32, 300:600], # 14-31 (1st)
# self.attribute[32:34, 900:], # 32, 33 (3rd)
# self.attribute[34:89, 300:600], # 34-88 (1st)
# self.attribute[89, 900:], # 89 (3rd)
# self.attribute[90:, 300:600])) # 90-100 (1st)
####################################################################################
# Average Experiments
# Case: Average action with best object
#print("Average action with best object.")
#self.attribute = (self.action + self.best_obj) / 2
# Case: Average(Class, 1 object, 2 object, 3 object)
#print("Average(Class, 1 object, 2 object, 3 object)")
#self.attribute = (self.action + self.obj1 + self.obj2 + self.obj3) / 4
# Case: Average(1 object, 2 object, 3 object)
#print("Average(1 object, 2 object, 3 object)")
#self.attribute = (self.obj1 + self.obj2 + self.obj3) / 3
# Case: Average(Class, 1 object)
#print("Average(Class, 1 object)")
#self.attribute = (self.action + self.obj1) / 2
# Case: Average(Class, 2 object)
#print ("Average(Class, 2 object)")
#self.attribute = (self.action + self.obj2) / 2
# Case: Average(Class, 3 object)
#print ("Average(Class, 3 object)")
#self.attribute = (self.action + self.obj3) / 2
# Case: Average(Class, 1 object, 2 object)
#print ("Average(Class, 1 object, 2 object)")
#self.attribute = (self.action + self.obj1 + self.obj2) / 3
# Case: Average(Class, 1 object, 3 object)
#print ("Average(Class, 1 object, 3 object)")
#self.attribute = (self.action + self.obj1 + self.obj3) / 3
# Case: Average(Class, 2 object, 3 object)
print ("Average(Class, 2 object, 3 object)")
self.attribute = (self.action + self.obj2 + self.obj3) / 3
####################################################################################
# Case 1: Replace action wv with object wv (300d)
#print("replace action wv with 1st object")
#self.attribute = self.attribute[:, 300:600]
#print("replace action wv with 2nd object")
#self.attribute = self.attribute[:, 600:900]
#print("replace action wv with 3rd object")
#self.attribute = self.attribute[:, 900:]
################################################################################################
# Case 2: Append 1 object (600d)
#print("append 1st obj.")
#self.attribute = self.attribute[:, :600]
#print("append 2nd obj.")
# a[:,2:6] - including index 2 and excluding index 6
#self.attribute = torch.hstack((self.attribute[:, :300], self.attribute[:, 600:900]))
#print("append 3rd obj.")
#self.attribute = torch.hstack((self.attribute[:, :300], self.attribute[:, 900:]))
################################################################################################
# Case 3: Append 2 objects (900d)
# 1st + 2nd
#print("append 2 objects (1st + 2nd)")
#self.attribute = self.attribute[:, :900]
# 1st + 3rd
#print("append 2 objects (1st + 3rd)")
#self.attribute = torch.hstack((self.attribute[:, :600], self.attribute[:, 900:]))
# 2nd + 3rd
#print("append 2 objects (2nd + 3rd)")
#self.attribute = torch.hstack((self.attribute[:, :300], self.attribute[:, 600:]))
################################################################################################
print(self.attribute.shape)
self.attribute /= self.attribute.pow(2).sum(1).sqrt().unsqueeze(1).expand(self.attribute.size(0),
self.attribute.size(1))
elif opt.avg_wv:
self.attribute = torch.from_numpy(matcontent['att_all'].T).float()
self.action_wv = torch.from_numpy(matcontent['att'].T).float()
self.object1_wv = self.attribute[:, 300:600]
self.object2_wv = self.attribute[:, 600:900]
self.object3_wv = self.attribute[:, 900:]
# Case 1: avg action and 1st object
print("Averaging action and 1st object semantics")
self.attribute = torch.add(self.action_wv, self.object1_wv) / 2
print(self.attribute.shape)
else:
print("without object semantics:")
self.attribute = torch.from_numpy(matcontent['att'].T).float()
self.attribute /= self.attribute.pow(2).sum(1).sqrt().unsqueeze(1).expand(self.attribute.size(0),
self.attribute.size(1))
print(self.attribute.shape)
else:
print("Wrong semantics. In UCF101 splits file, att means word2vec and origin_att means attributes.")
elif opt.dataset == "hmdb51":
# load visual features for HMDB51
matcontent = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/"
+ opt.dataset + "_" + opt.action_embedding + ".mat")
# matcontent = sio.loadmat(opt.dataroot + "/" + opt.image_embedding_path + "/" + opt.image_embedding + ".mat")
feature = matcontent['features'].T
label = matcontent['labels'].astype(int).squeeze() - 1
# load action dataset splits and semantics
matcontent = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + opt.dataset + "_semantics/" +
"split_1/" + "att_splits.mat")
# matcontent = sio.loadmat(opt.dataroot + "/" + opt.dataset + "/" + opt.class_embedding + "_splits.mat")
trainval_loc = matcontent['trainval_loc'].squeeze() - 1
train_loc = matcontent['train_loc'].squeeze() - 1
val_unseen_loc = matcontent['val_loc'].squeeze() - 1
#test_seen_loc = matcontent['test_seen_loc'].squeeze() - 1
test_unseen_loc = matcontent['test_unseen_loc'].squeeze() - 1
if opt.class_embedding == "wv":
self.attribute = torch.from_numpy(matcontent['att'].T).float()
self.attribute /= self.attribute.pow(2).sum(1).sqrt().unsqueeze(1).expand(self.attribute.size(0),
self.attribute.size(1))
else:
print("Wrong semantics. In HMDB51 splits file, att means word2vec.")
else:
print("Wrong dataset!")
'''
# use above codes
if opt.manual_att:
print("Using manual_att")
m_att = torch.from_numpy(np.load(opt.dataroot + "/ucf101_i3d/ucf101_manual_att.npy")).float()
m_att /= m_att.pow(2).sum(1).sqrt().unsqueeze(1).expand(101,m_att.size(1))
self.attribute = m_att
'''
if not opt.validation:
print("Disable cross validation mode")
if opt.preprocessing:
print('Preprocessing (MinMaxScaler)...')
if opt.standardization:
print('Standardization...')
scaler = preprocessing.StandardScaler()
# scaler_att = preprocessing.StandardScaler()
else:
scaler = preprocessing.MinMaxScaler()
# scaler_att = preprocessing.MinMaxScaler()
_train_feature = scaler.fit_transform(feature[train_loc])
#_test_seen_feature = scaler.transform(feature[test_seen_loc])
_test_unseen_feature = scaler.transform(feature[test_unseen_loc])
self.train_feature = torch.from_numpy(_train_feature).float()
mx = self.train_feature.max()
self.train_feature.mul_(1 / mx)
self.train_label = torch.from_numpy(label[train_loc]).long()
self.test_unseen_feature = torch.from_numpy(_test_unseen_feature).float()
self.test_unseen_feature.mul_(1 / mx)
self.test_unseen_label = torch.from_numpy(label[test_unseen_loc]).long()
#self.test_seen_feature = torch.from_numpy(_test_seen_feature).float()
#self.test_seen_feature.mul_(1 / mx)
#self.test_seen_label = torch.from_numpy(label[test_seen_loc]).long()
# Scaled and transformed (0,1) attributes (bce: binary class embedding)
# self.bce_att = opt.bce_att
# select either binary class embedding or norm class embedding for attributes
# if opt.bce_att:
# temp_att = torch.from_numpy(scaler_att.fit_transform(self.original_att)).float()
# else:
# temp_att = torch.from_numpy(scaler_att.fit_transform(self.attribute)).float()
# mx_att = temp_att.max()
# temp_att.mul_(1/mx)
# self.bce_attribute = temp_att
# self.bce_attribute_norm = self.bce_attribute/self.bce_attribute.pow(2).sum(1).sqrt().unsqueeze(1).expand(self.attribute.size(0), self.attribute.size(1))
else:
self.train_feature = torch.from_numpy(feature[train_loc]).float()
self.train_label = torch.from_numpy(label[train_loc]).long()
self.test_unseen_feature = torch.from_numpy(feature[test_unseen_loc]).float()
self.test_unseen_label = torch.from_numpy(label[test_unseen_loc]).long()
#self.test_seen_feature = torch.from_numpy(feature[test_seen_loc]).float()
#self.test_seen_label = torch.from_numpy(label[test_seen_loc]).long()
else:
print("Enable cross validation mode")
self.train_feature = torch.from_numpy(feature[train_loc]).float()
self.train_label = torch.from_numpy(label[train_loc]).long()
self.test_unseen_feature = torch.from_numpy(feature[val_unseen_loc]).float()
self.test_unseen_label = torch.from_numpy(label[val_unseen_loc]).long()
self.seenclasses = torch.from_numpy(np.unique(self.train_label.numpy()))
self.unseenclasses = torch.from_numpy(np.unique(self.test_unseen_label.numpy()))
self.ntrain = self.train_feature.size()[0]
#self.ntest_seen = self.test_seen_feature.size()[0]
self.ntest_unseen = self.test_unseen_feature.size()[0]
self.ntrain_class = self.seenclasses.size(0)
self.ntest_class = self.unseenclasses.size(0)
self.train_class = self.seenclasses.clone()
self.allclasses = torch.arange(0, self.ntrain_class + self.ntest_class).long()
self.train_mapped_label = map_label(self.train_label, self.seenclasses)
def next_seen_batch(self, seen_batch):
idx = torch.randperm(self.ntrain)[0:seen_batch]
batch_feature = self.train_feature[idx]
batch_label = self.train_label[idx]
batch_att = self.attribute[batch_label]
# batch_bce_att = self.bce_attribute[batch_label]
return batch_feature, batch_att # batch_bce_att