-
Notifications
You must be signed in to change notification settings - Fork 1
/
data.py
97 lines (86 loc) · 3.89 KB
/
data.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
import torch
from torch.utils.data import Dataset
import pandas as pd
import cv2
from tqdm import tqdm
import numpy as np
class CASMECombinedDataset(Dataset):
def __init__(self,path = '.',
img_sz = 128,
calculate_strain = False,
raw_img = False,
initialized_df = None):
if initialized_df is None:
print('Initializing CASME Combined Dataset...')
self.df = pd.read_csv(path + '/' + 'combined_3class.csv')
self.df['OpticalFlow'] = None
self.path = path
self.img_sz = img_sz
for idx,row in tqdm(self.df.iterrows(),ascii = '='):
prefix = self.__get_prefix(row)
onset = self.__read_img(
prefix + str(row['Onset']) + '.jpg'
)
apex = self.__read_img(
prefix + str(row['Apex']) + '.jpg'
)
if raw_img:
self.df.at[idx,'OpticalFlow'] = np.vstack(
(np.expand_dims(onset.astype(np.float32) / 255, axis = 0),np.expand_dims(apex.astype(np.float32) / 255, axis = 0))
)
else:
self.df.at[idx,'OpticalFlow'] = self.__calc_optical_flow(onset,apex)
if calculate_strain:
self.df.at[idx,'OpticalFlow'] = self.__append_optical_strain(self.df.at[idx,'OpticalFlow'])
self.df.at[idx,'Class'] = {'negative':0,'positive':1,'surprise':'2'}[row['Class']]
else:
self.df = initialized_df
def __get_prefix(self,row):
sub_sample = row['Subject'] + '/' + row['Sample'] + '/'
if row['Dataset'] == 'casme1':
return self.path + '/casme1_cropped/' + sub_sample + 'reg_' + row['Sample'] + '-'
elif row['Dataset'] == 'casme2':
return self.path + '/casme2_cropped/' + sub_sample + 'reg_img'
elif row['Dataset'] == 'casme^2':
return self.path + '/casme^2_cropped/' + sub_sample + 'img'
def __read_img(self,name):
return cv2.cvtColor(
cv2.resize(
cv2.imread(name,cv2.IMREAD_COLOR),
(self.img_sz,self.img_sz),
interpolation = cv2.INTER_CUBIC
),
cv2.COLOR_BGR2GRAY
)
def __calc_optical_flow(self,onset,apex):
return np.array(
cv2.optflow.DualTVL1OpticalFlow_create().calc(onset,apex,None)
).transpose((2,0,1))
def __append_optical_strain(self,flow):
ux = cv2.Sobel(flow[0],cv2.CV_32F,1,0)
uy = cv2.Sobel(flow[0],cv2.CV_32F,0,1)
vx = cv2.Sobel(flow[1],cv2.CV_32F,1,0)
vy = cv2.Sobel(flow[1],cv2.CV_32F,0,1)
strain = np.sqrt(ux * ux + uy * uy + 0.5 * (vx + uy) * (vx + uy))
return np.concatenate((flow,strain.reshape(1,self.img_sz,self.img_sz)),axis = 0)
def __len__(self):
return len(self.df)
def __getitem__(self,idx):
return self.df.at[idx,'OpticalFlow'],int(self.df.at[idx,'Class'])
class LOSOGenerator():
def __init__(self,dataset):
self.data = dataset
self.subjects = self.data.df[['Dataset','Subject']].drop_duplicates().reset_index()
self.idx = 0
def __iter__(self):
return self
def __next__(self):
if self.idx == len(self.subjects):
raise StopIteration
ds,sub = self.subjects.at[self.idx,'Dataset'],self.subjects.at[self.idx,'Subject']
self.idx += 1
train_df = self.data.df[(self.data.df.Dataset != ds) | (self.data.df.Subject != sub)] \
.reset_index(drop=True)
test_df = self.data.df[(self.data.df.Dataset == ds) & (self.data.df.Subject == sub)] \
.reset_index(drop=True)
return CASMECombinedDataset(initialized_df = train_df),CASMECombinedDataset(initialized_df = test_df)