/
smic_processing.py
240 lines (186 loc) · 8 KB
/
smic_processing.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
import os
import sys
import cv2
import numpy as np
import pandas as pd
import face_recognition
import matplotlib.pyplot as plt
from tqdm import tqdm
data_root = '/home/ubuntu/Datasets/MEGC/smic/HS_long/SMIC_HS_E/'
smic_annotation_file = 'datasets/SMIC-HS-E_annotation.xlsx'
label_dict = {'negative': 0, 'positive': 1, 'surprise': 2}
def get_clip_frame_paths(subject, filename, on_frame_idx, off_frame_idx):
frame_paths = []
subject = 's{}'.format(str(subject).zfill(2))
dir_path = os.path.join(data_root, subject, filename)
for idx in range(on_frame_idx, off_frame_idx + 1):
idx = str(idx).zfill(6)
frame_path = os.path.join(dir_path, 'image{}.jpg'.format(idx))
if not os.path.exists(frame_path):
print('Fail to locate file', frame_path)
raise Exception('The value of path was: {}'.format(frame_path))
frame_paths.append(frame_path)
return frame_paths
def detect_lmks(frame):
lmks = face_recognition.face_landmarks(frame)
return lmks[0]
def get_cell(img, cell_location):
point1, point2 = cell_location
cell = img[point1[1]:point2[1], point1[0]:point2[0]]
return cell
def get_cell_locations(lmks):
def get_rect(center, width):
point1 = np.array(center) - int(width / 2)
point2 = np.array(center) + int(width / 2)
return tuple(point1), tuple(point2)
cells = {}
cell_width = int((lmks['top_lip'][6][0] - lmks['top_lip'][0][0]) / 2)
key = 'top_lip'
points = np.array(lmks[key])
left_lip_rect = get_rect(points[0], cell_width)
right_lip_rect = get_rect(points[6], cell_width)
cells['left_lip'] = left_lip_rect
cells['right_lip'] = right_lip_rect
key = 'chin'
point = lmks[key][int(len(lmks[key]) / 2)]
rect_point1 = (point[0] - int(cell_width / 2), point[1] - cell_width)
rect_point2 = (point[0] + int(cell_width / 2), point[1])
chin_rect = (rect_point1, rect_point2)
cells['chin_rect'] = chin_rect
key = 'nose_tip'
point = lmks[key][0]
left_nose_rect_point1 = (point[0] - cell_width, left_lip_rect[0][1] - cell_width)
left_nose_rect_point2 = (point[0], left_lip_rect[0][1])
left_nose_rect = (left_nose_rect_point1, left_nose_rect_point2)
cells['left_nose'] = left_nose_rect
point = lmks[key][4]
right_nose_rect_point1 = (point[0], right_lip_rect[0][1] - cell_width)
right_nose_rect_point2 = (point[0] + cell_width, right_lip_rect[0][1])
right_nose_rect = (right_nose_rect_point1, right_nose_rect_point2)
cells['right_nose'] = right_nose_rect
key = 'left_eye'
point = lmks[key][0]
left_eye_rect_point1 = (point[0] - cell_width, int(point[1] - cell_width / 2))
left_eye_rect_point2 = (point[0], int(point[1] + cell_width / 2))
left_eye_rect = (left_eye_rect_point1, left_eye_rect_point2)
cells['left_eye'] = left_eye_rect
key = 'right_eye'
point = lmks[key][3]
right_eye_rect_point1 = (point[0], int(point[1] - cell_width / 2))
right_eye_rect_point2 = (point[0] + cell_width, int(point[1] + cell_width / 2))
right_eye_rect = (right_eye_rect_point1, right_eye_rect_point2)
cells['right_eye'] = right_eye_rect
left_point = lmks['left_eyebrow'][2]
right_point = lmks['right_eyebrow'][2]
center_point = (int((left_point[0] + right_point[0]) / 2),
int((left_point[1] + right_point[1]) / 2))
center_eyebrow_rect = get_rect(center_point, cell_width)
cells['center_eyebrow'] = center_eyebrow_rect
left_rect_point1 = (int(center_point[0] - cell_width * 3 / 2),
int(center_point[1] - cell_width / 2))
left_rect_point2 = (int(center_point[0] - cell_width * 1 / 2),
int(center_point[1] + cell_width / 2))
left_eyebrow_rect = (left_rect_point1, left_rect_point2)
cells['left_eyebrow'] = left_eyebrow_rect
right_rect_point1 = (int(center_point[0] + cell_width * 1 / 2),
int(center_point[1] - cell_width / 2))
right_rect_point2 = (int(center_point[0] + cell_width * 3 / 2),
int(center_point[1] + cell_width / 2))
right_eyebrow_rect = (right_rect_point1, right_rect_point2)
cells['right_eyebrow'] = right_eyebrow_rect
return cells, cell_width
def compute_cell_difference(cell_t, cell_onset, cell_offset, cell_epsilon):
numerator = (np.abs(cell_t - cell_onset) + 1.0)
denominator = (np.abs(cell_t - cell_epsilon) + 1.0)
difference = numerator / denominator
numerator = (np.abs(cell_t - cell_offset) + 1.0)
difference1 = numerator / denominator
# difference = difference + difference1
return difference.mean()
def compute_cell_features(frame_t, on_frame, off_frame, frame_epsilon):
lmks = detect_lmks(frame_t)
cell_locations, cell_width = get_cell_locations(lmks)
cell_differences = {}
frame_t = frame_t.astype(np.float32)
on_frame = on_frame.astype(np.float32)
off_frame = off_frame.astype(np.float32)
frame_epsilon = frame_epsilon.astype(np.float32)
for key in cell_locations:
cell_location = cell_locations[key]
cell_t = get_cell(frame_t, cell_location)
cell_onset = get_cell(on_frame, cell_location)
cell_offset = get_cell(off_frame, cell_location)
cell_epsilon = get_cell(frame_epsilon, cell_location)
cell_difference = compute_cell_difference(cell_t, cell_onset, cell_offset, cell_epsilon)
cell_differences[key] = cell_difference
return cell_differences
def find_apex_frame_of_clip(frame_paths):
epsilon = 1
on_frame = cv2.imread(frame_paths[0], cv2.IMREAD_GRAYSCALE)
off_frame = cv2.imread(frame_paths[-1], cv2.IMREAD_GRAYSCALE)
features = []
for i in range(epsilon, len(frame_paths)):
frame_t = cv2.imread(frame_paths[i], cv2.IMREAD_GRAYSCALE)
frame_epsilon = cv2.imread(frame_paths[i - epsilon], cv2.IMREAD_GRAYSCALE)
current_features = compute_cell_features(frame_t, on_frame, off_frame, frame_epsilon)
feature = 0
for key in current_features:
feature += current_features[key]
feature = feature / len(current_features)
features.append(feature)
padding = [0.0] * epsilon
features = np.array(padding + features)
apex_frame_idx = features.argmax()
apex_frame_path = frame_paths[apex_frame_idx]
return apex_frame_path, features, apex_frame_idx
def draw_avg_plot(features, pred_apex_idx, data, clip_name):
x = list(range(len(features)))
plt.plot(x, features)
plt.axvline(x=pred_apex_idx, label='pred apex idx at={}'.format(pred_apex_idx), c='red')
plt.legend()
plt.savefig('plots/{}/{}.png'.format(data, clip_name))
plt.clf();
plt.cla();
plt.close();
def on_all_smic_clips():
smic = pd.read_excel(smic_annotation_file)
labels = []
apex_frame_indices = []
on_frame_paths = []
off_frame_paths = []
apex_frame_paths = []
samples = zip(list(smic['Subject']),
list(smic['Filename']),
list(smic['OnsetF']),
list(smic['OffsetF']),
list(smic['Emotion']))
with tqdm(total=158) as progress_bar:
for subject, filename, on_frame_idx, off_frame_idx, emotion in samples:
# Get all ME paths of a clip
clip_frame_paths = get_clip_frame_paths(subject, filename, on_frame_idx, off_frame_idx)
# Find apex frame paths
apex_frame_path, features, apex_relative_idx = find_apex_frame_of_clip(clip_frame_paths)
draw_avg_plot(features, apex_relative_idx, 'smic', filename)
on_frame_paths.append(clip_frame_paths[0])
off_frame_paths.append(clip_frame_paths[-1])
apex_frame_paths.append(apex_frame_path)
apex_frame_idx = int(apex_frame_path.split('/')[-1].split('.')[0].replace('image', ''))
apex_frame_indices.append(apex_frame_idx)
# Label
labels.append(label_dict[emotion])
progress_bar.update(1)
# Save data_to_csv file
data_dict = {'data' : ['smic'] * len(labels),
'subject' : list(smic['Subject']),
'clip' : list(smic['Filename']),
'label' : labels,
'onset_frame' : list(smic['OnsetF']),
'apex_frame' : apex_frame_indices,
'offset_frame' : list(smic['OffsetF']),
'onset_frame_path': on_frame_paths,
'apex_frame_path' : apex_frame_paths,
'off_frame_path' : off_frame_paths}
smic_data = pd.DataFrame.from_dict(data_dict)
smic_data.to_csv('datasets/smic_apex.csv', header=True, index=None)
if __name__ == '__main__':
on_all_smic_clips()