/
ferplus_dataset.py
251 lines (213 loc) · 8.59 KB
/
ferplus_dataset.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
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
import numpy as np
import time
from math import ceil
import random
import cv2
import sys
import os
from dataset_tools import enclosing_square, add_margin, cut
import keras
from tqdm import tqdm
from dataset_tools import DataGenerator
from dataset_tools import _readcsv
from dataset_tools import linear_balance_illumination, mean_std_normalize, equalize_hist
from six.moves import cPickle as pickle
from threading import Lock
NUM_CLASSES = 8
def _fer_to_img(line_arr):
data = [int(x) for x in line_arr[1].split(' ')]
data = np.array(data, dtype=np.uint8)
data = np.reshape(data, (48,48,1))
#data = cv2.cvtColor(data, cv2.COLOR_GRAY2RGB)
return data
def cntk_filtering(data):
# remove outlier votes
data = np.array([int(x) for x in data])
outliers = data<=1
data[outliers] = 0
totalvotes = np.sum(data)
# remove examples from class 9 or 10
hardlabel = np.argmax(data)
if hardlabel == 8 or hardlabel==9:
return True, None
# remove examples with more than two winners
maxvotes = np.max(data)
winners = data==maxvotes
nwinners = np.sum(winners)
if nwinners > 2:
return True, None
# remove examples where the winners have <=50% of all votes
numwinnervotes = nwinners*maxvotes
if numwinnervotes <= 0.5*totalvotes:
return True, None
# return normalized
data = data.astype(float)/totalvotes
return False, data[0:NUM_CLASSES]
def _load_ferplus(meta, images, partition):
data = []
n_discarded=0
n_discarded_cntk=0
for n,d in enumerate(tqdm(meta[1:])):
if d[0]==partition:
if d[1]=='':
n_discarded+=1
else:
try:
drop, labels = cntk_filtering(d[2:])
#n = int(d[1][3:10])
#print(d[1],n)
if not drop:
example={
'img': _fer_to_img(images[n+1]),
'label': labels,
'roi': (0,0,48,48)
}
if np.max(example['img'])==np.min(example['img']):
print('Warning, blank image!')
else:
data.append(example)
else:
n_discarded_cntk+=1
except (IndexError) as e:
print('not found: ' +str(e))
pass
print("Data loaded. %d samples (%d+%d discarded)" % (len(data), n_discarded, n_discarded_cntk) )
return data
class FerPlusDataset:
def __init__(self, partition='train', csvdata='FERPlus/fer2013.csv', csvmeta='FERPlus/fer2013new.csv', target_shape=(48,48,1), augment=True, custom_augmentation=None, preprocessing='full_normalization', debug_max_num_samples=None):
if partition.startswith('train'):
partition='Training'
elif partition.startswith('val'):
partition='PrivateTest'
elif partition.startswith('test'):
partition='PublicTest'
else:
raise Exception("unknown partition")
self.target_shape = target_shape
self.custom_augmentation = custom_augmentation
self.augment = augment
self.gen = None
self.preprocessing = preprocessing
print('Loading data...')
cache_file_name = '%s.%s%s.cache'%(csvdata,partition, '.'+str(debug_max_num_samples) if debug_max_num_samples is not None else '')
try:
with open(cache_file_name, 'rb') as f:
self.data = pickle.load(f)
self.data = self.data[:debug_max_num_samples]
print("Data loaded. %d samples, from cache" % (len(self.data)) )
except FileNotFoundError:
meta = _readcsv(csvmeta)
images = _readcsv(csvdata)
print('csv read complete: %d, %d.' %(len(meta), len(images)))
self.data = _load_ferplus(meta, images, partition)
with open(cache_file_name, 'wb') as f:
pickle.dump(self.data, f)
def get_num_samples(self):
return self.data.shape[0]
def get_num_classes(self):
return NUM_CLASSES
def get_generator(self, batch_size=64):
if self.gen is None:
self.gen = DataGenerator(self.data, self.target_shape, with_augmentation=self.augment, custom_augmentation=self.custom_augmentation, batch_size=batch_size, preprocessing=self.preprocessing)
return self.gen
CLASS_LABELS = ['neutral','happiness','surprise','sadness','anger','disgust','fear','contempt','unknown','NF']
def draw_emotion(y, w,h):
EMOTIONS=CLASS_LABELS
COLORS = [(120,120,120), (50,50,255), (0,255,255), (255,0,0), (0,0,140), (0,200,0), (42,42,165), (100,100,200), (170,170,170), (80,80,80)]
emotionim = np.zeros((w,h,3), dtype=np.uint8)
barh = h//len(EMOTIONS)
MAXEMO = np.sum(y)
for i,yi in enumerate(y):
#print((EMOTIONS[i], yi))
p1,p2 = (0,i*barh), (int(yi*w//MAXEMO), (i+1)*20)
cv2.rectangle(emotionim, p1,p2, COLORS[i], cv2.FILLED)
cv2.putText(emotionim, "%s: %.1f" % (EMOTIONS[i], yi), (0,i*20+14), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (255, 255, 255))
return emotionim
def findRelevantFace(objs, W,H):
mindistcenter = None
minobj = None
for o in objs:
cx = o['roi'][0] + (o['roi'][2]/2)
cy = o['roi'][1] + (o['roi'][3]/2)
distcenter = (cx-(W/2))**2 + (cy-(H/2))**2
if mindistcenter is None or distcenter < mindistcenter:
mindistcenter = distcenter
minobj = o
return minobj
def top_left(f):
return (f['roi'][0], f['roi'][1])
def bottom_right(f):
return (f['roi'][0]+f['roi'][2], f['roi'][1]+f['roi'][3])
def dump_ds(dataset,name,h5file):
dt=dataset
tgen = dt.get_generator(1)
Xt=[]
Yt=[]
for batch in tqdm(tgen):
Xt.append(batch[0])
Yt.append(batch[1])
Xt=np.squeeze(np.array(Xt))
print(Xt.shape, Xt.dtype)
Yt=np.squeeze(np.array(Yt))
dsfile.create_dataset('X'+name, data=Xt)
dsfile.create_dataset('Y'+name, data=Yt)
def test1():
print('Training')
dt = FerPlusDataset(target_shape=(224,224,3), preprocessing='full_normalization', debug_max_num_samples=None)
print('Now generating from training set')
gen = dt.get_generator()
i=0
while True:
print(i)
i+=1
for batch in tqdm(gen):
for im, identity in zip(batch[0], batch[1]):
identity = np.argmax(identity)
facemax = np.max(im)
facemin = np.min(im)
im = (255*( (im - facemin) / (facemax - facemin) )).astype(np.uint8)
cv2.imshow('ferplus image', im)
cv2.waitKey(0)
if '__main__' == __name__:
test1()
def dumpall():
print('Training')
dt = FerPlusDataset(target_shape=(224,224,3), preprocessing='full_normalization', debug_max_num_samples=None, augment=False)
print('Validation')
dv = FerPlusDataset('val',target_shape=(224,224,3), augment=False, preprocessing='full_normalization', debug_max_num_samples=None)
print('Test')
dp = FerPlusDataset('test',target_shape=(224,224,3), augment=False, preprocessing='full_normalization', debug_max_num_samples=None)
print('Now generating from training set')
import h5py
dsfile=h5py.File('ds.h5','w')
dump_ds(dt, 't',dsfile)
dump_ds(dv, 'v',dsfile)
dump_ds(dp, 's',dsfile)
dsfile.close()
'''
from face_detector import FaceDetector
from face_aligner import FaceAligner
fd = FaceDetector(min_confidence=0.8)
al = FaceAligner()
for batch in gen:
for x,y in zip(batch[0], batch[1]):
window = np.zeros((400,200,3), dtype=np.uint8)
x = ((x*127)+127).clip(0,255).astype(np.uint8)
if len(x.shape)<=2 or x.shape[2]==1:
x = cv2.cvtColor(x, cv2.COLOR_GRAY2BGR)
window[0:200,0:200,:] = x
faces = fd.detect(x)
#f = findRelevantFace(faces, x.shape[1], x.shape[0])
for f in faces:
print(f['roi'], f['confidence'])
br = list(bottom_right(f))
if br[1] > 200: br[1]=200
if br[0] > 200: br[0]=200
cv2.rectangle(window, top_left(f), tuple(br), (255, 255, 0), 2)
window[200:400,0:200,:] = draw_emotion(y,200,200)
cv2.imshow('im', window)
cv2.waitKey(0)
print("------------------------")
print("------------------------")
print("------------------------")
'''