-
Notifications
You must be signed in to change notification settings - Fork 0
/
demo.py
145 lines (121 loc) · 5.76 KB
/
demo.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
import skvideo.io
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
from torchvision import models, transforms
from torch.autograd import Variable
from skimage.metrics import structural_similarity as ssim
import scipy
import argparse
weight=[1/30,1,5,1]
non_uni=[0.367,0.633]
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"-n", "--name", type=str,
default="example.mp4",
help="input video path",
)
parser.add_argument(
"-i", "--info", type=str,
default="example.csv",
help="video information"
)
parser.add_argument(
"-q", "--qos", type=str,
default="height",
help="QoS feature representing quality"
)
parser.add_argument(
"-d", "--dir", type=str,
default="example.mat",
help="feature save dictionary"
)
parser.add_argument(
"-s", "--sample", type=int,
default=30,
help="global sampling intensity"
)
args = parser.parse_args()
videoName=args.name
videoIndex=args.info
dirSave=args.dir
QoS=args.qos
videoData = skvideo.io.vread(videoName,as_grey=True)
videoInfo = pd.read_csv(videoIndex)
start_index=np.zeros(5,dtype='int')
end_index=np.zeros(5,dtype='int')
start_index[0]=int(np.round(videoInfo['framerate'][0]*videoInfo['rebuffering_duration'][0]))
end_index[0]=start_index[0]+int(np.round(videoInfo['framerate'][0]*videoInfo['chunk_duration'][0]))
for clips in range(1,5):
start_index[clips]=end_index[clips-1]+int(np.round(videoInfo['framerate'][clips]*videoInfo['rebuffering_duration'][clips]))
end_index[clips]=start_index[clips]+int(np.round(videoInfo['framerate'][clips]*videoInfo['chunk_duration'][clips]))
reward_QoS=np.zeros(5)
reward_content=np.zeros(25)
penalty_QoS=np.zeros(2)
penalty_content=np.zeros(4)
##################### reward QoS #####################
######################################################
for clips in range(0,5):
reward_QoS[clips]=videoInfo[QoS][clips]
######################################################
##################### reward content #################
######################################################
gru = nn.GRU(input_size=4, hidden_size=4,
num_layers=2, # gru层数
batch_first=False, # 默认参数 True:(batch, seq, feature) False：True:( seq,batch, feature),
bidirectional=False, # 默认参数
)
#hid = torch.randn(2 * 1, 1, 4)
hid = torch.Tensor([[[ 0.3818, -0.0660, -0.0082, -0.4698]],[[-0.2710, 0.1756, 0.3829, 0.4570]]])
for clips in range(0,5):
start_frame=start_index[clips]
end_frame=end_index[clips]
resnet50_feature_extractor = models.resnet50(pretrained=True) # 导入ResNet50的预训练模型
resnet50_feature_extractor.conv1=nn.Conv2d(1, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False) # 黑白
resnet50_feature_extractor.fc = nn.Linear(2048, 16) # 最后16个元素
#torch.nn.init.eye(resnet50_feature_extractor.fc.weight) # 将二维tensor初始化为单位矩阵
transform1 = transforms.Compose([
transforms.Scale(256), # 缩放
transforms.CenterCrop(224), # 中心裁剪
transforms.ToTensor()] # 转换成Tensor
)
now_frame=start_frame
while now_frame<end_frame:
x = Variable(torch.unsqueeze(transform1(transforms.ToPILImage()(videoData[now_frame,:,:,:])), dim=0).float(), requires_grad=False)
y = resnet50_feature_extractor(x)
y = y.data.numpy()
inter_frame=args.sample
if now_frame*2<(start_frame+end_frame):
inter_frame=inter_frame*non_uni[0]
else:
inter_frame=inter_frame*non_uni[1]
now_frame=now_frame+int(inter_frame*len(non_uni))
output, hid = gru(torch.Tensor([[[np.max(y),np.min(y),np.mean(y),np.std(y)]]]), hid)
reward_content[clips*5:(clips*5+4)]=output.data.numpy()[0][0]
imgComplex=x.data.numpy()[0,0,:,:]
tmp=np.zeros([16,16])
ave=scipy.linalg.block_diag(tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp,tmp)/16
mat1=ave*imgComplex
mat2=imgComplex*ave
mat3=(mat1+mat2)/2
reward_content[clips*5+4]=min(np.mean(abs(imgComplex-mat1)),np.mean(abs(imgComplex-mat2)),np.mean(abs(imgComplex-mat3)))
######################################################
##################### penalty QoS ####################
######################################################
penalty_QoS[0]=videoInfo['rebuffering_duration'][0]
penalty_QoS[1]=np.mean(videoInfo['rebuffering_duration'][1:5])
######################################################
##################### penalty content ################
######################################################
c1=0.5
c2=100
for clips in range(0,4):
rebuffer=videoInfo['rebuffering_duration'][clips]
switch=max(videoInfo[QoS][clips]-videoInfo[QoS][clips+1],0)
structure=ssim(videoData[end_index[clips]-5,:,:,0],videoData[start_index[clips+1]+5,:,:,0])
penalty_content[clips]=(1+rebuffer/c1)*(1+switch/c2)*structure
######################################################
scipy.io.savemat(dirSave, mdict={'all': np.hstack((reward_QoS*weight[0], reward_content*weight[1], penalty_QoS*weight[2], penalty_content*weight[3]))})
print("The feature of " + videoName + " is saved")