Repository navigation
Expand file tree
/
Copy pathutils.py
More file actions
299 lines (274 loc) · 12 KB
/
Copy pathutils.py
File metadata and controls
299 lines (274 loc) · 12 KB
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
import pandas as pd
import cv2
import os
import numpy as np
from tqdm import tqdm
import pickle
from sklearn.model_selection import StratifiedKFold
from tqdm import tqdm
from PIL import Image, ImageDraw
import random
from keras.utils import to_categorical
import face_recognition
def to_pkl(content,pkl_file):
"""
save data in pkl format
@param content:data
@param pkl_file: save path
"""
with open(pkl_file, 'wb') as f:
pickle.dump(content, f)
f.close()
def TV_L1_os2(prev,curr,bound=15):
"""
TV_L1 method
@param prev: Array, the first frame image array
@param curr: Array, the current frame image array
@param bound: Int, the bound
@return: u,v,os; optical flow images
"""
prev=cv2.cvtColor(prev, cv2.COLOR_BGR2GRAY)
curr=cv2.cvtColor(curr, cv2.COLOR_BGR2GRAY)
TVL1=cv2.optflow.DualTVL1OpticalFlow_create()
flow = TVL1.calc(prev, curr, None)
assert flow.dtype == np.float32
flow = (flow + bound) * (255.0 / (2 * bound))
flow = np.round(flow).astype(int)
flow[flow >= 255] = 255
flow[flow <= 0] = 0
u=flow[:, :, 0]
v=flow[:, :, 1]
(x,y)=u.shape
u_x=u-u[:,[0]+[i for i in range(0,y-1)]]
v_y=v-v[[0]+[i for i in range(0,x-1)],:]
u_y=u-u[[0]+[i for i in range(0,x-1)],:]
v_x=v-v[:,[0]+[i for i in range(0,y-1)]]
os_=np.sqrt(u_x**2+v_y**2+0.5*(u_y+v_x)**2)
u=u[:, : ,np.newaxis]
v=v[:, : ,np.newaxis]
os_=os_[:, : ,np.newaxis]
return u,v,os_
def detectFace_loc(path):
"""
Given a picture, it returns the top, right, bottom, left coordinates of the face in the picture
@param path: image path
@return: top, right, bottom, left coordinates of the face in the picture
"""
image = cv2.imread(path)
face_locations = face_recognition.face_locations(image)
top, right, bottom, left = face_locations[0]
return top, right, bottom, left
def generate_samm_trainset():
"""
genarate the train data and the train label of SAMM,apex frame denotes 1,and other frames randomly sampled denotes 0.
@return: the train set,Dataframe
"""
#handle csv
df=pd.read_csv('data/casme_groundTruth.csv')
df=df[['fold','sub_fold','Onset','Apex','Offset','type']]
df=df.loc[df['type']=='macro-expression']
df['last']=df['Offset']-df['Onset']
df=df.loc[df['last']>=10]
#generate the train set of SAMM dataset
data=pd.DataFrame()
for index, row in df.iterrows():
if row['fold']=='none' or row['Apex']==0:
continue
if row['Apex']>=1000:
apex_path=os.path.join('img'+str(row['Apex'])+'.jpg')
else:
apex_path=os.path.join('img'+str(row['Apex']).zfill(3)+'.jpg')
first_path=os.path.join('img'+str(1).zfill(3)+'.jpg')
if row['Onset']-20>=1000:
path1=os.path.join('img'+str(row['Onset']-20)+'.jpg')
else:
path1=os.path.join('img'+str(row['Onset']-20).zfill(3)+'.jpg')
path2=os.path.join('img'+str(25).zfill(3)+'.jpg')
path3=os.path.join('img'+str(50).zfill(3)+'.jpg')
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":apex_path,"label":1},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path1,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path2,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path3,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
tmp=pd.read_csv('data/casme_face_loc.csv')
for index, row in tmp.iterrows():
if len(df.loc[(df['fold']==row['fold'])&(df['sub_fold']==row['sub_fold'])])==0:
first_path=os.path.join('img'+str(1).zfill(3)+'.jpg')
path1=os.path.join('img'+str(25).zfill(3)+'.jpg')
path2=os.path.join('img'+str(50).zfill(3)+'.jpg')
path3=os.path.join('img'+str(75).zfill(3)+'.jpg')
path4=os.path.join('img'+str(100).zfill(3)+'.jpg')
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path1,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path2,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path3,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"fold":row['fold'],"sub_fold":row['sub_fold'],"onset":first_path,"frame":path4,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
data=data.drop_duplicates()
return data
def load_samm_train_data():
"""
load train set of SAMM dataset
@return: u,v:List,dynamic image; img:List, static image;
label:List,label list; fold_idx: 5-fold cross-validation trn_idx and val_idx
"""
trn_u_path='data/samm_train_u.pkl'
trn_v_path='data/samm_train_v.pkl'
trn_img_path='data/samm_train_img.pkl'
trn_label_path='data/samm_train_label.pkl'
trn_fold_path='data/samm_train_fold_idx.pkl'
add=0
if os.path.exists(trn_u_path) and os.path.exists(trn_v_path):
#if pkl exists, directly read
return pd.read_pickle(trn_u_path),pd.read_pickle(trn_v_path),pd.read_pickle(trn_img_path),pd.read_pickle(trn_label_path),pd.read_pickle(trn_fold_path)
else:
#else generate pkl,and return data
prefix_path='data/Cas(me)^2' #Put the data of casme2 in this path
macro=generate_samm_trainset()
u,v,img,label=[],[],[],[]
#Read data in the dataFrame line by line
for i in tqdm(macro.index):
try:
onset_path=os.path.join(prefix_path,macro.loc[i]['fold'],macro.loc[i]['sub_fold'],macro.loc[i]['onset'])
frame_path=os.path.join(prefix_path,macro.loc[i]['fold'],macro.loc[i]['sub_fold'],macro.loc[i]['frame'])
#generate the (u,v,f)
t,r,b,l=detectFace_loc(onset_path)
onset=cv2.imread(onset_path)
onset = onset[t-add:b+add,l-add:r+add]
onset=cv2.resize(onset,(160,160))
prev=cv2.imread(frame_path)
prev = prev[t-add:b+add,l-add:r+add]
prev=cv2.resize(prev,(160,160))
tmp_u,tmp_v,tmp_os=TV_L1_os2(prev,onset)
u.append(tmp_u)
v.append(tmp_v)
img.append(prev)
label.append(macro.loc[i]['label'])
except Exception as e:
print(e)
print(macro.loc[i]['frame'])
u=np.array(u)
v=np.array(v)
img=np.array(img)
label=np.array(label)
# Random disruption
index = [i for i in range(len(u))]
random.shuffle(index)
u = u[index]
v=v[index]
img=img[index]
label= label[index]
#print(u.shape)
#print(img.shape)
split=[]
skf = StratifiedKFold(n_splits=5,random_state=2020,shuffle=True)
for trn_idx,val_idx in skf.split(u,label):
split.append((trn_idx,val_idx))
#to pkl
to_pkl(u,trn_u_path)
to_pkl(v,trn_v_path)
to_pkl(img,trn_img_path)
to_pkl(label,trn_label_path)
to_pkl(split,trn_fold_path)
return u,v,img,label,split
def generate_casme_trainset():
"""
genarate the train data and the train label of CAS(ME)2,apex frame denotes 1,and other frames randomly sampled denotes 0.
@return: the train set,Dataframe
"""
data=pd.DataFrame()
df=pd.read_csv('data/SAMM_groundTruth.csv')
#df=df.loc[df['Type']=='Macro']
for index, row in df.iterrows():
if row['Apex']>10000 or row['Onset']>10000:
print(row)
apex_path=row['Subject']+'_'+str(row['Apex']).zfill(4)+'.jpg'
first_path=row['Subject']+'_'+str(1).zfill(4)+'.jpg'
path1=row['Subject']+'_'+str(row['Onset']-20).zfill(4)+'.jpg'
path2=row['Subject']+'_'+str(25).zfill(4)+'.jpg'
path3=row['Subject']+'_'+str(50).zfill(4)+'.jpg'
if row['Type']=='Macro':
new= pd.DataFrame({"Subject":row["Subject"],"onset":first_path,"frame":apex_path,"label":1},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"Subject":row["Subject"],"onset":first_path,"frame":path1,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"Subject":row["Subject"],"onset":first_path,"frame":path2,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
new= pd.DataFrame({"Subject":row["Subject"],"onset":first_path,"frame":path3,"label":0},index=["0"])
data=data.append(new,ignore_index=True)
#print(len(data))
#print(data.head())
data=data.drop_duplicates()
#print(len(data))
#print(data['label'].value_counts())
return data
def load_casme_train_data():
"""
load train set of CASME dataset
@return: u,v:List,dynamic image; img:List, static image;
label:List,label list; fold_idx:5-fold cross-validation trn_idx and val_idx
"""
trn_u_path='data/casme_train_u.pkl'
trn_v_path='data/casme_train_v.pkl'
trn_img_path='data/casme_train_img.pkl'
trn_label_path='data/casme_train_label.pkl'
trn_fold_path='data/casme_train_fold_idx.pkl'
add=0
if os.path.exists(trn_u_path) and os.path.exists(trn_v_path):
#if pkl exists, directly read
return pd.read_pickle(trn_u_path),pd.read_pickle(trn_v_path),pd.read_pickle(trn_img_path),pd.read_pickle(trn_label_path),pd.read_pickle(trn_fold_path)
else:
#else generate pkl,and return data
prefix_path='data/SAMM' #Put the data of samm in this path
macro=generate_casme_trainset() #use the Cross dataset assessment,use the SAMM datset to be the train set of CAS(ME)2
u,v,img,label=[],[],[],[]
#Read data in the dataFrame line by line
for i in tqdm(macro.index):
try:
onset_path=os.path.join(prefix_path,macro.loc[i]['Subject'],macro.loc[i]['onset'])
frame_path=os.path.join(prefix_path,macro.loc[i]['Subject'],macro.loc[i]['frame'])
#generate the (u,v,f)
t,r,b,l=detectFace_loc(onset_path)
onset=cv2.imread(onset_path)
onset = onset[t-add:b+add,l-add:r+add]
onset=cv2.resize(onset,(160,160))
prev=cv2.imread(frame_path)
prev = prev[t-add:b+add,l-add:r+add]
prev=cv2.resize(prev,(160,160))
tmp_u,tmp_v,tmp_os=TV_L1_os2(prev,onset)
u.append(tmp_u)
v.append(tmp_v)
img.append(prev)
label.append(macro.loc[i]['label'])
except Exception as e:
print(e)
print(macro.loc[i]['frame'])
u=np.array(u)
v=np.array(v)
img=np.array(img)
label=np.array(label)
# Random disruption
index = [i for i in range(len(u))]
random.shuffle(index)
u = u[index]
v=v[index]
img=img[index]
label= label[index]
#print(u.shape)
#print(img.shape)
split=[]
skf = StratifiedKFold(n_splits=5,random_state=2020,shuffle=True)
for trn_idx,val_idx in skf.split(u,label):
split.append((trn_idx,val_idx))
#to pkl
to_pkl(u,trn_u_path)
to_pkl(v,trn_v_path)
to_pkl(img,trn_img_path)
to_pkl(label,trn_label_path)
to_pkl(split,trn_fold_path)
return u,v,img,label,split