-
Notifications
You must be signed in to change notification settings - Fork 0
/
new_preprocessing.py
172 lines (146 loc) · 6.41 KB
/
new_preprocessing.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
import glob
import numpy as np
import pandas as pd
import os
import json
import ast
def get_openpose_from_json(video_id, start_frame, window_size):
"""
:param video_id: the id of video you want to get openpose
:param start_frame: the frame when the bite action starts
:param window_size: the window size of each sample in seconds
"""
path = '/home/aa2375/social-dining/data/openpose/' + str(video_id)
frame_list = os.listdir(path)
frame_list.sort()
print("len:", len(frame_list))
b_x, b_y, b_c = [], [], []
lh_x, lh_y, lh_c = [], [], []
rh_x, rh_y, rh_c = [], [], []
f_x, f_y, f_c = [], [], []
for i in range(start_frame - 30 * window_size, start_frame):
with open(path + '/' + frame_list[i]) as f:
pose_frame = json.load(f)
if not pose_frame['people']: # if no body detected:
body_joints = np.empty(25 * 3)
body_joints[:] = np.nan
face_joints = np.empty(70 * 3)
face_joints[:] = np.nan
else:
body_joints = pose_frame['people'][0]['pose_keypoints_2d']
face_joints = pose_frame['people'][0]['face_keypoints_2d']
a = 0
while a < (len(body_joints)):
b_x.append(body_joints[a])
b_y.append(body_joints[a + 1])
b_c.append(body_joints[a + 1])
a += 3
a = 0
while a < (len(face_joints)):
f_x.append(face_joints[a])
f_y.append(face_joints[a + 1])
f_c.append(face_joints[a + 1])
a += 3
body_op_x = np.array(b_x).reshape(window_size * 30, -1)
body_op_y = np.array(b_y).reshape(window_size * 30, -1)
body_op_c = np.array(b_c).reshape(window_size * 30, -1)
# remove keypoints on legs and feet
body_op_x = np.delete(body_op_x, [10, 11, 13, 14, 18, 19, 20, 21, 22, 23, 24], 1)
body_op_y = np.delete(body_op_y, [10, 11, 13, 14, 18, 19, 20, 21, 22, 23, 24], 1)
body_op_c = np.delete(body_op_c, [10, 11, 13, 14, 18, 19, 20, 21, 22, 23, 24], 1)
face_op_x = np.array(f_x).reshape(window_size * 30, -1)
face_op_y = np.array(f_y).reshape(window_size * 30, -1)
face_op_c = np.array(f_c).reshape(window_size * 30, -1)
# sample = np.concatenate((np.concatenate((np.concatenate((np.concatenate((np.concatenate(
# (np.concatenate((np.concatenate((body_op_x, body_op_y), axis=1), lhand_op_x), axis=1), lhand_op_y),
# axis=1), rhand_op_x), axis=1), rhand_op_y), axis=1), face_op_x), axis=1), face_op_y), axis=1)
body_sample = np.concatenate([body_op_x, body_op_y], axis=1)
face_sample = np.concatenate([face_op_x, face_op_y], axis=1)
return np.array(body_sample).astype(float), np.array(face_sample).astype(float)
def mapping_op_sample_to_elan_label(elan_label_path, out_folder='positive_samples'):
elan_label = pd.read_csv(elan_label_path, index_col=0)
op_sample_list = []
op_sample_1_list = []
op_sample_2_list = []
vid_1_list = []
vid_2_list = []
for i in range(len(elan_label)):
vid_main = elan_label['Name'].iloc[i]
print(vid_main)
start_frame = elan_label['Start Frame'].iloc[i]
print(start_frame)
# this determines the left and right people!
if vid_main[-1] == '1':
vid_1_list.append(vid_main[0:-1]+str(2))
vid_2_list.append(vid_main[0:-1] + str(3))
elif vid_main[-1] == '2':
vid_1_list.append(vid_main[0:-1] + str(1))
vid_2_list.append(vid_main[0:-1] + str(3))
elif vid_main[-1] == '0':
continue
else:
vid_1_list.append(vid_main[0:-1] + str(1))
vid_2_list.append(vid_main[0:-1] + str(2))
op_sample = get_openpose_from_json(vid_main, start_frame, 6)
op_name = f"data/{out_folder}/" + str(i).zfill(5) + '_main.npy'
np.save(op_name, op_sample)
op_sample_list.append(op_name)
op_1_sample = get_openpose_from_json(vid_1_list[i], start_frame, 6)
op_1_name = f"data/{out_folder}/" + str(i).zfill(5) + '_1.npy'
np.save(op_1_name, op_1_sample)
op_sample_1_list.append(op_1_name)
op_2_sample = get_openpose_from_json(vid_2_list[i], start_frame, 6)
op_2_name = f"data/{out_folder}/" + str(i).zfill(5) + '_2.npy'
np.save(op_2_name, op_2_sample)
op_sample_2_list.append(op_2_name)
elan_label['video_id_1'] = vid_1_list
elan_label['video_id_2'] = vid_2_list
elan_label['op_main'] = op_sample_list
elan_label['op_1'] = op_sample_1_list
elan_label['op_2'] = op_sample_2_list
elan_label.to_csv(f"data/{out_folder}_with_op.csv")
def interpolate_gaze_headpose():
df = pd.read_csv("/Users/tongwu/Downloads/rt_gene_feats.csv", usecols=['name', 'gaze'])
l = []
for i in range(1, len(df) - 1):
pre = df.iloc[i-1, 0][-5:]
cur = df.iloc[i, 0][-5:]
if int(pre) != int(cur) - 1 and df.iloc[i-1, 0][:3] ==df.iloc[i, 0][:3]:
diff = int(cur) - int(pre)
for j in range(diff):
should = int(pre) + j
l.append(df.iloc[i-1, 0][:-5] + str(should).zfill(5))
else:
l.append(df.iloc[i-1, 0])
new_col = pd.DataFrame({"new_name": l})
df = df.set_index("name")
new_col = new_col.set_index("new_name")
df_new = new_col.join(df)
df_new = df_new.interpolate()
print()
def get_gaze_from_csv(video_id, start_frame, window_size):
"""
:param video_id: the id of video you want to get openpose
:param start_frame: the frame when the bite action starts
:param window_size: the window size of each sample in seconds
"""
gaze_df = pd.read_csv(f"data/rt_gene/{video_id}.csv")
gaze_df = gaze_df.set_index('name')
start_id = video_id + '_' + str(start_frame).zfill(5)
end_id = video_id + '_' + str(start_frame + 30 * window_size).zfill(5)
data = gaze_df.loc[start_id : end_id]
gaze = np.array(data['gaze'])
head_pose = np.array(data['headpose'])
for i in range(len(gaze)):
gaze[i] = np.array(ast.literal_eval(gaze[i]))
head_pose[i] = np.array(ast.literal_eval(head_pose[i]))
return np.array(gaze), np.array(head_pose)
# load positive csv
# add openposes to it
mapping_op_sample_to_elan_label('data/positive_labels.csv', 'positive_samples')
# add gazes to it
# add speaking to it
# add count to it
# load negative csv
# add openposes to it
mapping_op_sample_to_elan_label('data/negative_labels.csv', 'negative_samples')