-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathfully_supervised_main.py
198 lines (154 loc) · 7.63 KB
/
fully_supervised_main.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
from __future__ import print_function
import os
import time
import torch
import torch.nn as nn
import numpy as np
import random
import torch.optim as optim
from torch.optim.lr_scheduler import StepLR
from dataloader import AVE_Fully_Dataset
from fully_model import psp_net
from measure import compute_acc, AVPSLoss
from Optim import ScheduledOptim
import warnings
warnings.filterwarnings("ignore")
import argparse
import pdb
parser = argparse.ArgumentParser(description='Fully supervised AVE localization')
# data
parser.add_argument('--model_name', type=str, default='PSP', help='model name')
parser.add_argument('--dir_video', type=str, default="./data/visual_feature.h5", help='visual features')
parser.add_argument('--dir_audio', type=str, default='./data/audio_feature.h5', help='audio features')
parser.add_argument('--dir_labels', type=str, default='./data/right_labels.h5', help='labels of AVE dataset')
parser.add_argument('--dir_order_train', type=str, default='./data/train_order.h5', help='indices of training samples')
parser.add_argument('--dir_order_val', type=str, default='./data/val_order.h5', help='indices of validation samples')
parser.add_argument('--dir_order_test', type=str, default='./data/test_order.h5', help='indices of testing samples')
parser.add_argument('--nb_epoch', type=int, default=300, help='number of epoch')
parser.add_argument('--batch_size', type=int, default=128, help='number of batch size')
parser.add_argument('--save_epoch', type=int, default=5, help='number of epoch for saving models')
parser.add_argument('--check_epoch', type=int, default=5, help='number of epoch for checking accuracy of current models during training')
parser.add_argument('--LAMBDA', type=float, default=100, help='weight for balancing losses')
parser.add_argument('--threshold', type=float, default=0.099, help='key-parameter for pruning process')
parser.add_argument('--trained_model_path', type=str, default=None, help='pretrained model')
parser.add_argument('--train', action='store_true', default=False, help='train a new model')
FixSeed = 123
random.seed(FixSeed)
np.random.seed(FixSeed)
torch.manual_seed(FixSeed)
torch.cuda.manual_seed(FixSeed)
def train(args, net_model, optimizer):
AVEData = AVE_Fully_Dataset(video_dir=args.dir_video, audio_dir=args.dir_audio, label_dir=args.dir_labels,
order_dir=args.dir_order_train, batch_size=args.batch_size, status='train')
nb_batch = AVEData.__len__() // args.batch_size
print('nb_batch:', nb_batch)
epoch_l = []
best_val_acc = 0
best_test_acc = 0
best_epoch = 0
for epoch in range(args.nb_epoch):
net_model.train()
epoch_loss = 0
epoch_loss_cls = 0
epoch_loss_avps = 0
n = 0
start = time.time()
SHUFFLE_SAMPLES = True
for i in range(nb_batch):
audio_inputs, video_inputs, labels, segment_label_batch, segment_avps_gt_batch = AVEData.get_batch(i, SHUFFLE_SAMPLES)
SHUFFLE_SAMPLES = False
audio_inputs = audio_inputs.cuda()
video_inputs = video_inputs.cuda()
labels = labels.cuda()
segment_label_batch = segment_label_batch.cuda()
segment_avps_gt_batch = segment_avps_gt_batch.cuda()
net_model.zero_grad()
fusion, out_prob, cross_att = net_model(audio_inputs, video_inputs, args.threshold) # shape:
# out_prob: [bs, 10, 29], score_max: [bs, 29]
loss_cls = nn.CrossEntropyLoss()(out_prob.permute(0, 2, 1), segment_label_batch) # segment_label_batch: [bs, 10]
loss_avps = AVPSLoss(cross_att, segment_avps_gt_batch)
loss = loss_cls + args.LAMBDA * loss_avps
epoch_loss += loss.cpu().data.numpy()
epoch_loss_cls += loss_cls.cpu().data.numpy()
epoch_loss_avps += loss_avps.cpu().data.numpy()
loss.backward()
optimizer.step_lr()
n = n + 1
SHUFFLE_SAMPLES = True
if (epoch+1) % 60 == 0 and epoch < 170:
optimizer.update_lr()
end = time.time()
epoch_l.append(epoch_loss)
labels = labels.cpu().data.numpy()
x_labels = out_prob.cpu().data.numpy()
acc = compute_acc(labels, x_labels, nb_batch)
print("=== Epoch {%s} lr: {%.6f} | Loss: [{%.4f}] loss_cls: [{%.4f}] | loss_frame: [{%.4f}] | training_acc {%.4f}" \
% (str(epoch), optimizer._optimizer.param_groups[0]['lr'], (epoch_loss) / n, epoch_loss_cls/n, epoch_loss_avps/n, acc))
if epoch % args.save_epoch == 0 and epoch != 0:
val_acc = val(args, net_model)
print('val accuracy:', val_acc, 'epoch=', epoch)
if val_acc >= best_val_acc:
best_val_acc = val_acc
print('best val accuracy:', best_val_acc)
print('best val accuracy: {} ***************************************'.format(best_val_acc))
# torch.save(net_model, model_name + "_" + str(epoch) + "_fully.pt")
if epoch % args.check_epoch == 0 and epoch != 0:
test_acc = test(args, net_model)
print('test accuracy:', test_acc, 'epoch=', epoch)
if test_acc >= best_test_acc:
best_test_acc = test_acc
best_epoch = epoch
print('best test accuracy: {} ======================================='.format(best_test_acc))
torch.save(net_model, model_name + "_" + str(epoch) + "_fully.pt")
print('[best val accuracy]: ', best_val_acc)
print('[best test accuracy]: ', best_test_acc)
def val(args, net_model):
net_model.eval()
AVEData = AVE_Fully_Dataset(video_dir=args.dir_video, audio_dir=args.dir_audio, label_dir=args.dir_labels,
order_dir=args.dir_order_val, batch_size=402, status='val')
nb_batch = AVEData.__len__()
audio_inputs, video_inputs, labels, _, _ = AVEData.get_batch(0)
audio_inputs = audio_inputs.cuda()
video_inputs = video_inputs.cuda()
labels = labels.cuda()
fusion, out_prob, cross_att = net_model(audio_inputs, video_inputs, args.threshold)
labels = labels.cpu().data.numpy()
x_labels = out_prob.cpu().data.numpy()
acc = compute_acc(labels, x_labels, nb_batch)
print('[val]acc: ', acc)
return acc
def test(args, net_model, model_path=None):
if model_path is not None:
net_model = torch.load(model_path)
print(">>> [Testing] Load pretrained model from " + model_path)
net_model.eval()
AVEData = AVE_Fully_Dataset(video_dir=args.dir_video, audio_dir=args.dir_audio, label_dir=args.dir_labels,
order_dir=args.dir_order_test, batch_size=402, status='test')
nb_batch = AVEData.__len__()
audio_inputs, video_inputs, labels, _, _, = AVEData.get_batch(0)
audio_inputs = audio_inputs.cuda()
video_inputs = video_inputs.cuda()
labels = labels.cuda()
fusion, out_prob, cross_att = net_model(audio_inputs, video_inputs, args.threshold)
labels = labels.cpu().data.numpy()
x_labels = out_prob.cpu().data.numpy()
acc = compute_acc(labels, x_labels, nb_batch)
print('[test]acc: ', acc)
return acc
if __name__ == "__main__":
args = parser.parse_args()
print("args: ", args)
# model and optimizer
model_name = args.model_name
if model_name == "PSP":
net_model = psp_net(128, 512, 128, 29)
else:
raise NotImplementedError
net_model.cuda()
optimizer = optim.Adam(net_model.parameters(), lr=1e-3)
optimizer = ScheduledOptim(optimizer)
if args.train:
train(args, net_model, optimizer)
else:
test_acc = test(args, net_model, model_path=args.trained_model_path)
print("[test] accuracy: ", test_acc)