-
Notifications
You must be signed in to change notification settings - Fork 1
/
predict_hdrmsssim.py
200 lines (163 loc) · 5.83 KB
/
predict_hdrmsssim.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
import argparse
import glob
import itertools
import os
import re
import sys
from os.path import join
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import sklearn
from scipy import stats
from scipy.stats import pearsonr, spearmanr
from sklearn.ensemble import RandomForestRegressor
# from joblib import Parallel, delayed
from sklearn.model_selection import GridSearchCV, train_test_split
from sklearn.neural_network import MLPRegressor
from sklearn.preprocessing import MinMaxScaler, StandardScaler
from sklearn.svm import SVR
from training import *
def remove_extensions(file_name):
if file_name.endswith('.yuv'):
return file_name[:-4]
elif file_name.endswith('.mp4'):
return file_name[:-4]
else:
return file_name
def process_string(s):
s = s.replace('[', '')
s = s.replace(']', '')
s = s.replace(' ', ' ')
s = s.replace(' ', ' ')
s = s.replace('\n', '')
sp = re.split(' |,', s)
sp = [value for value in sp if value != '']
l = [eval(i) for i in sp]
return l
def process_dlm(df, method='by_level'):
# if the method is by band, then returns only four levels, otherwise returns the entire bands
if method == 'by_level':
nums2 = df.iloc[:, 1]
nums2 = [process_string(i) for i in nums2]
nums2 = np.array(nums2)
nums2 = nums2.mean(axis=0)
return nums2
else:
nums3 = df.iloc[:, 2]
nums3 = [process_string(i) for i in nums3]
nums3 = np.array(nums3)
nums3 = nums3.mean(axis=0)
return nums3
def process_vif(df, method='by_level'):
# if the method is by band, then returns only four levels, otherwise returns the entire bands
nums2 = df.iloc[:, 1]
nums2 = [process_string(i) for i in nums2]
nums2 = np.array(nums2)
nums2 = nums2
nums3 = df.iloc[:, 2]
nums3 = [process_string(i) for i in nums3]
nums3 = np.array(nums3)
res = nums2.reshape(-1, 4, 2).sum(axis=2) / \
(nums3.reshape(-1, 4, 2).sum(axis=2)+0.0001)
res = res.mean(axis=0)
return res
def read_features(file_path):
files = glob.glob(join(file_path, '*.csv'))
vnames = []
feats = []
if files == []:
print('em')
if file_path.find('dlm') >= 0:
for f in files:
df = pd.read_csv(f, index_col=0)
feats_1vid = process_dlm(df)
vname = os.path.basename(f)[:-4]
vnames.append(vname)
feats.append(feats_1vid)
else:
for f in files:
df = pd.read_csv(f, index_col=0)
feats_1vid = process_vif(df)
vname = os.path.basename(f)[:-4]
vnames.append(vname)
feats.append(feats_1vid)
features = pd.DataFrame(np.array(feats))
features['video'] = vnames
return features
def conbine_texp_features(feats_pth):
bright_pth = join(feats_pth, f'hdrdlmnew/dlm_local_m_exp_None/ycbcr_0')
dlm_feats_non = read_two_exp_features(bright_pth)
bright_pth = join(feats_pth, f'hdrvifnew/vif_local_m_exp_None/ycbcr_0')
vif_feats_non = read_two_exp_features(bright_pth)
non_feats = vif_feats_non.merge(dlm_feats_non, on='video')
return non_feats
def read_two_exp_features(pth1):
files = glob.glob(join(pth1, '*.csv'))
vnames = []
feats = []
if files == []:
print('em')
if pth1.find('dlm') >= 0:
for f in files:
df = pd.read_csv(f, skiprows=1, header=None, index_col=0)
feats_1vid = process_dlm(df.iloc[:, [0, 1, 2]])
vname = os.path.basename(f)[:-4]
vnames.append(vname)
feats.append(feats_1vid)
else:
for f in files:
df = pd.read_csv(f, skiprows=1, header=None, index_col=0)
feats_1vid = process_vif(df.iloc[:, [0, 1, 2]])
vname = os.path.basename(f)[:-4]
vnames.append(vname)
feats.append(feats_1vid)
features1 = pd.DataFrame(np.array(feats))
features1['video'] = vnames
return features1
def predict(df, svr_pth, scaler_pth):
X_test = df.drop(['video'], axis=1)
scaler = StandardScaler()
svrfile = open(svr_pth, 'rb')
grid_svr = pickle.load(svrfile)
svrfile.close()
scalerfile = open(scaler_pth, 'rb')
scaler = pickle.load(scalerfile)
scalerfile.close()
X_test = scaler.transform(X_test)
predict = grid_svr.predict(X_test)
df['pred'] = predict
return df
def combine_ssim(feats_pth):
csv_files = glob.glob(join(feats_pth, 'hdrmsssimnew/*.csv'))
vnames = []
feats = []
for f in csv_files:
df = pd.read_csv(f)
vname = os.path.basename(f)[:-4]
vnames.append(vname)
# read the whole row as features
feats.append(df.values.reshape(-1))
features = pd.DataFrame(np.array(feats))
features['video'] = vnames
return features
argparser = argparse.ArgumentParser(
description='Predicting the video level features')
argparser.add_argument('feature_path', type=str,
help='Path to the folder containing the features')
argparser.add_argument('output_name', type=str,
help='Output name', default='predict.csv')
argparser.add_argument('--svr_name', type=str,
help='SVR name, to be used with CUSTOM model',default="./models/svr/msssim_svr.pkl")
argparser.add_argument('--scaler_name', type=str,
help='Scaler name, to be used with CUSTOM model', default="./models/scaler/msssim_scaler.pkl")
args = argparser.parse_args()
output_name = args.output_name
feats_pth = args.feature_path
scaler_name = os.path.abspath(args.scaler_name)
svr_name = os.path.abspath(args.svr_name)
ssim_features = combine_ssim(feats_pth)
nonlinear_features = conbine_texp_features(feats_pth)
feature = ssim_features.merge(nonlinear_features, on='video')
res = predict(feature, args.svr_name, args.scaler_name)
res.to_csv(output_name)