-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathhmdb51_att_lstm.py
315 lines (231 loc) · 10.9 KB
/
hmdb51_att_lstm.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
"""
LSTM model for action recognition
Author: Lili Meng [email protected], March 12th, 2018
"""
from __future__ import print_function
import sys
import os
import math
import shutil
import random
import tempfile
import unittest
import traceback
import torch
import torch.utils.data
import torch.cuda
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from torch.utils.serialization import load_lua
from tensorboardX import SummaryWriter
import torchvision.transforms as transforms
import argparse
import numpy as np
import time
from PIL import Image
use_cuda = True
class Action_Att_LSTM(nn.Module):
def __init__(self, input_size, hidden_size, output_size, seq_len):
super(Action_Att_LSTM, self).__init__()
self.hidden_size = hidden_size
self.lstm = nn.LSTM(input_size, hidden_size)
self.fc = nn.Linear(hidden_size, output_size)
self.fc_attention = nn.Linear(hidden_size, seq_len)
self.input_size = input_size
def forward(self, input_x):
#output = self.batch_norm_input(input_x)
# input_x shape [batch_size, 2048, 15]
# before doing LSTM, we need to swap channels as the LSTM accepts [temporal, batch_size, feature_size]
batch_size = input_x.shape[0]
tmp=input_x.transpose(0,2).transpose(1,2)
hidden = (self.init_hidden(batch_size), self.init_hidden(batch_size))
output, hidden = self.lstm(tmp, hidden)
output1 = output
#Only the last dimension of LSTM output is used, that is output[14]=output[-1] here
att_weight = self.fc_attention(output[-1])
att_weight = F.softmax(att_weight, dim =1)
weighted_output = torch.sum(output*att_weight.transpose(0, 1).unsqueeze(dim=2),
dim = 0)
return scores, att_weight
def init_hidden(self, batch_size):
result = Variable(torch.zeros(1, batch_size, self.hidden_size))
if use_cuda:
return result.cuda()
else:
return result
def lr_scheduler(optimizer, epoch_num, init_lr = 0.001, lr_decay_epochs=10):
"""Decay learning rate by a factor of 0.1 every lr_decay_epochs.
"""
using_cyclic_lr = False
if using_cyclic_lr == True:
eta_min = 5e-6
eta_max = 5e-4
lr = eta_min + 0.5 * (eta_max - eta_min) * (1 + np.cos(epoch_num/FLAGS.max_epoch * np.pi))
else:
lr = init_lr *(0.1**(epoch_num//lr_decay_epochs))
if epoch_num % lr_decay_epochs == 0:
print("Learning rate changed to be : {}".format(lr))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
return optimizer
def train(batch_size,
train_data,
train_label,
model,
model_optimizer,
criterion):
"""
a training sample which goes through a single step of training.
"""
loss = 0
model_optimizer.zero_grad()
model_input = (train_data).view(batch_size, -1, FLAGS.num_segments)
model_input = model_input.cuda()
logits, att_weight = model.forward(model_input)
loss += criterion(logits, train_label)
att_reg = F.relu(att_weight[:, :-2] * att_weight[:, 2:] - att_weight[:, 1:-1].pow(2)).sqrt().mean()
factor = FLAGS.hp_reg_factor
if FLAGS.use_regularizer:
loss += factor*att_reg
loss.backward()
model_optimizer.step()
final_loss = loss.data[0]/batch_size
corrects = (torch.max(logits, 1)[1].view(train_label.size()).data == train_label.data).sum()
train_accuracy = 100.0 * corrects/batch_size
return final_loss, train_accuracy, att_weight
def test_step(batch_size,
batch_x,
batch_y,
model):
test_data_batch = batch_x.view(batch_size, -1, FLAGS.num_segments).cuda()
test_logits, test_att_weight = model(test_data_batch)
corrects = (torch.max(test_logits, 1)[1].view(batch_y.size()).data == batch_y.data).sum()
test_accuracy = 100.0 * corrects/batch_size
return test_logits, test_accuracy, test_att_weight
def main():
torch.manual_seed(1234)
dataset_name = FLAGS.dataset
maxEpoch = FLAGS.max_epoch
num_segments = FLAGS.num_segments
train_data = np.load("./saved_features/train_hmdb51_features.npy")
train_label = np.load("./saved_features/train_hmdb51_labels.npy")
train_name = np.load("./saved_features/train_hmdb51_names.npy")
test_data = np.load("./saved_features/test_hmdb51_features.npy")
test_label = np.load("./saved_features/test_hmdb51_labels.npy")
test_name = np.load("./saved_features/test_hmdb51_names.npy")
train_data = torch.from_numpy(train_data)
train_data = train_data.transpose(1,2)
train_label = torch.from_numpy(train_label)
test_data = torch.from_numpy(test_data)
test_data = test_data.transpose(1,2)
test_label = torch.from_numpy(test_label)
transform = transforms.Compose([
#transforms.Scale([224, 224]),
transforms.ToTensor()])
replace_with_random_noise_end = False
replace_with_random_noise_middle = False
noisy_train = torch.randn(train_data.shape[0], train_data.shape[1], 5)
noisy_test = torch.randn(test_data.shape[0], test_data.shape[1], 5)
if replace_with_random_noise_end == True:
train_data = torch.cat((train_data[:,:,0:10], noisy_train),2)
test_data = torch.cat((test_data[:,:,0:10], noisy_test), 2)
if replace_with_random_noise_middle == True:
train_data1 = torch.cat((train_data[:,:,0:5], noisy_train), 2)
train_data = torch.cat((train_data1, train_data[:,:,10:15]),2)
test_data1 = torch.cat((test_data[:,:,0:5], noisy_test), 2)
test_data = torch.cat((test_data1, test_data[:,:,10:15]),2)
print("train_data.shape: ", train_data.shape)
print("train_label.shape: ", train_label.shape)
replicate_frames_5_times = False
if replicate_frames_5_times == True:
train_data = np.repeat(train_data, 5, axis=2)
train_label = np.repeat(train_label, 5, axis=0)
test_data = np.repeat(test_data, 5, axis=2)
test_label = np.repeat(test_label, 5, axis=0)
lstm_action = Action_Att_LSTM(input_size=2048, hidden_size=512, output_size=51, seq_len=FLAGS.num_segments).cuda()
model_optimizer = torch.optim.Adam(lstm_action.parameters(), lr=5e-4)
criterion = nn.CrossEntropyLoss()
best_test_accuracy = 0
num_step_per_epoch_train = train_data.shape[0]//FLAGS.train_batch_size
num_step_per_epoch_test = test_data.shape[0]//FLAGS.test_batch_size
log_dir = os.path.join('./4class_hmdb51_tensorboard', 'reg_factor_'+str(FLAGS.hp_reg_factor)+time.strftime("_%b_%d_%H_%M", time.localtime()))
if not os.path.exists(log_dir):
os.makedirs(log_dir)
writer = SummaryWriter(log_dir)
for epoch_num in range(maxEpoch):
model_optimizer = lr_scheduler(optimizer = model_optimizer, epoch_num=epoch_num, init_lr = 5e-4, lr_decay_epochs=100)
lstm_action.train()
avg_train_accuracy = 0
permutation = torch.randperm(train_data.shape[0])
for i in range(0, train_data.shape[0], FLAGS.train_batch_size):
indices = permutation[i:i+FLAGS.train_batch_size]
train_batch_x, train_batch_y, train_batch_name = train_data[indices], train_label[indices], train_name[indices]
train_batch_x = Variable(train_batch_x).cuda().float()
train_batch_y = Variable(train_batch_y).cuda().long()
print("train_batch_name[0:5] ", train_batch_name[0:5])
train_loss, train_accuracy, train_att_weight = train(FLAGS.train_batch_size, train_batch_x, train_batch_y, lstm_action, model_optimizer, criterion)
print("train_att_weight[0:5,:] ", train_att_weight[0:5,:])
print("train_att_weight.mean(dim=0) ",train_att_weight.mean(dim=0))
avg_train_accuracy+=train_accuracy
final_train_accuracy = avg_train_accuracy/num_step_per_epoch_train
print("epoch: "+str(epoch_num)+ " train accuracy: " + str(final_train_accuracy))
writer.add_scalar('with_att'+'_reg_factor_'+str(FLAGS.hp_reg_factor)+'_train_accuracy', final_train_accuracy, epoch_num)
save_train_file = FLAGS.dataset + "_numSegments"+str(FLAGS.num_segments)+"_regFactor_"+str(FLAGS.hp_reg_factor)+"_train_acc.txt"
with open(save_train_file, "a") as text_file:
print(f"{str(final_train_accuracy)}", file=text_file)
avg_test_accuracy = 0
lstm_action.eval()
for i in range(0, test_data.shape[0], FLAGS.test_batch_size):
test_indices = range(test_data.shape[0])[i: i+FLAGS.test_batch_size]
test_batch_x, test_batch_y, test_batch_name = test_data[test_indices], test_label[test_indices], test_name[test_indices]
test_batch_x = Variable(test_batch_x, volatile=True).cuda().float()
test_batch_y = Variable(test_batch_y, volatile=True).cuda().long()
test_logits, test_accuracy, test_att_weight = test_step(FLAGS.test_batch_size, test_batch_x, test_batch_y, lstm_action)
if i==30:
#print("test_batch_name[0:5,:]: ", test_batch_name[0:5, :])
if epoch_num == maxEpoch-1:
displayed_imgs_names = test_batch_name[-1,:]
displayed_imgs = [Image.open(frame_path).convert('RGB') for frame_path in displayed_imgs_names]
print("test_batch_name[-5:,:] ", test_batch_name[-5:,:])
print("test_att_weight[-5:,:] ", test_att_weight[-5:,:])
print("test_att_weight.mean(dim=0) ",test_att_weight.mean(dim=0))
for j in range(len(displayed_imgs)):
writer.add_image('Image_'+str(j)+'_att_weight_'+ str(test_att_weight.data.cpu().numpy()[-1][j]), transform(displayed_imgs[j]), epoch_num)
avg_test_accuracy+= test_accuracy
final_test_accuracy = avg_test_accuracy/num_step_per_epoch_test
print("epoch: "+str(epoch_num)+ " test accuracy: " + str(final_test_accuracy))
writer.add_scalar('with_att'+'_reg_factor_'+str(FLAGS.hp_reg_factor)+'_test_accuracy', final_test_accuracy, epoch_num)
save_test_file = FLAGS.dataset + "_numSegments"+str(FLAGS.num_segments)+"_regFactor_"+str(FLAGS.hp_reg_factor)+"_test_acc.txt"
with open(save_test_file, "a") as text_file1:
print(f"{str(final_test_accuracy)}", file=text_file1)
if final_test_accuracy > best_test_accuracy:
best_test_accuracy = final_test_accuracy
print('\033[91m' + "best test accuracy is: " +str(best_test_accuracy)+ '\033[0m')
# export scalar data to JSON for external processing
#writer.export_scalars_to_json("./saved_logs/all_scalars.json")
writer.close()
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', type=str, default='HMDB51',
help='dataset: "UCF101", "HMDB51"')
parser.add_argument('--train_batch_size', type=int, default=28,
help='train_batch_size: [64]')
parser.add_argument('--test_batch_size', type=int, default=30,
help='test_batch_size: [64]')
parser.add_argument('--max_epoch', type=int, default=20,
help='max number of training epoch: [20]')
parser.add_argument('--num_segments', type=int, default=15,
help='num of segments per video: [15]')
parser.add_argument('--use_changed_lr', dest='use_changed_lr',
help='not use change learning rate by default', action='store_true')
parser.add_argument('--use_regularizer', dest='use_regularizer',
help='use regularizer', action='store_false')
parser.add_argument('--hp_reg_factor', type=float, default=1,
help='multiply factor for regularization. [1]')
FLAGS, unparsed = parser.parse_known_args()
if len(unparsed) > 0:
raise Exception('Unknown arguments:' + ', '.join(unparsed))
print(FLAGS)
main()