-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_preprocessing.py
328 lines (262 loc) · 10.5 KB
/
data_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
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
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
import argparse
import os
from skimage import transform as tf
import numpy as np
import cv2
from tqdm import tqdm
import torch
from scipy.ndimage import gaussian_filter1d
from scipy import interpolate
import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision
parser = argparse.ArgumentParser()
parser.add_argument(
"--template-path",
type=str,
help="Template face image path",
default="reference.jpeg",
)
parser.add_argument(
"-i",
"--input-dir",
type=str,
help="Input directory path",
)
parser.add_argument(
"-o",
"--output-dir",
type=str,
help="Output directory path",
)
parser.add_argument(
"--exception-file",
type=str,
help="Exception file name",
default="exceptions.txt",
)
parser.add_argument(
"-fps",
type=int,
help="frame rate of video",
default=25,
)
parser.add_argument(
"--input-video-type",
type=str,
help="input video format",
default="mp4",
)
parser.add_argument(
"--no-landmarks",
type=bool,
help="Exclude landmark extraction",
default=False,
)
parser.add_argument(
"--overwrite",
type=bool,
help="If true overwrites existing data, if true skips that file",
default=False,
)
parser.add_argument(
"--test-video-output",
type=bool,
help="If true also writes video",
default=False,
)
def write_video(video_name,data):
frame_len = data.shape[0]
frame_width = int(data.shape[1])
frame_height = int(data.shape[2])
out_video = cv2.VideoWriter('{}_.avi'.format(video_name),
cv2.VideoWriter_fourcc('M','J','P','G'), 30, (frame_height,frame_width))
for i in range(frame_len):
frame = data[i]
out_video.write(frame)
out_video.release()
def load_video_frames(video_path):
cap = cv2.VideoCapture(video_path)
frame_list = []
while cap.isOpened():
ret, frame = cap.read()
if ret:
frame_list.append(frame)
else:
break
cap.release()
return frame_list
def detect_landmark(img,size):
mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=img)
detection_result = detector.detect(mp_image).face_landmarks
if not detection_result:
return None
else:
detection_result = detection_result[0]
landmark_array = np.zeros((len(all_landmark_coords),2))
for i in range(0,len(detection_result)):
if i in all_landmark_coords:
x = int(detection_result[i].x * size[1])
y = int(detection_result[i].y * size[0])
index = landmark_coords_dict[i]
landmark_array[index] = [x,y]
return landmark_array
# finds mean points of the left eye,right eye, and nose
def get_parts(landmark_array):
left_eye = landmark_array[20:36].mean(axis=0)
right_eye = landmark_array[36:52].mean(axis=0)
nose = landmark_array[128:135].mean(axis=0)
points = np.stack([left_eye, right_eye, nose])
return points
# draws a set of lines and connects dots between points in a list
def draw_line(image, color, thickness, coord_list, connect_start_end=False):
for i in range(1, len(coord_list)):
image = cv2.line(
image, tuple(coord_list[i - 1]), tuple(coord_list[i]), color, thickness
)
# whether first and last points should be connected (used for eye,lips and teeth)
if connect_start_end:
image = cv2.line(
image, tuple(coord_list[0]), tuple(coord_list[-1]), color, thickness
)
return image
# draws an entire face using landmarks and connecting them
def draw_face(preds, size, thickness=1,image=None):
if image is None:
image = np.zeros(size)
image = draw_line(image, (0, 0, 255), thickness, preds[92:128, :],connect_start_end=True) # face
image = draw_line(image, (255, 255, 0), thickness, preds[0:10, :], connect_start_end=True) # eye_brow1
image = draw_line(image, (0, 255, 255), thickness, preds[10:19, :], connect_start_end=True) # eye_brow2
image = draw_line(
image, (128, 0, 255), thickness, preds[20:36, :], connect_start_end=True
) # eye_1
image = draw_line(
image, (255, 128, 0), thickness, preds[36:52, :], connect_start_end=True
) # eye_2
image = draw_line(
image, (0, 255, 0), thickness, preds[72:92, :], connect_start_end=True
) # outer lips
image = draw_line(
image, (255, 255, 255), thickness, preds[52:72, :], connect_start_end=True
) # inner lips
image = draw_line(image, (255, 0, 0), thickness, preds[128:135, :]) # nose
image = image.astype("uint8")
return image
def draw_points(preds,size,image=None):
if image is None:
image = np.zeros(size)
for i in range(len(preds)):
image = cv2.circle(image, (preds[i,0],preds[i,1]), radius=1, color=(0, 255, 0), thickness=-1)
image = image.astype("uint8")
return image
args = parser.parse_args()
template_path = args.template_path
input_dir = args.input_dir
output_dir = args.output_dir
exception_file = args.exception_file
fps = args.fps
input_video_type = args.input_video_type
no_landmarks = args.no_landmarks
overwrite = args.overwrite
test_video_output = args.test_video_output
base_options = python.BaseOptions(model_asset_path='face_landmarker_v2_with_blendshapes.task')
options = vision.FaceLandmarkerOptions(base_options=base_options,
output_face_blendshapes=True,
output_facial_transformation_matrixes=True,
num_faces=1)
detector = vision.FaceLandmarker.create_from_options(options)
# list of corrupt files
corrupt_file_list = [
"1076_MTI_NEU_XX.mp4",
"1076_MTI_SAD_XX.mp4",
"1064_TIE_SAD_XX.mp4",
"1064_IEO_DIS_MD.mp4",
]
# Coordinates are thanks to :https://github.com/k-m-irfan/simplified_mediapipe_face_landmarks
Left_eyebrow = [70, 63, 105, 66, 107, 55, 65, 52, 53, 46]
Right_eyebrow = [300, 293, 334, 296, 336, 285, 295, 282, 283, 276]
Left_eye = [33, 246, 161, 160, 159, 158, 157, 173, 133, 155, 154, 153, 145, 144, 163, 7]
Right_eye = [263, 466, 388, 387, 386, 385, 384, 398, 362, 382, 381, 380, 374, 373, 390, 249]
Inner_Lip = [78, 191, 80, 81, 82, 13, 312, 311, 310, 415, 308, 324, 318, 402, 317, 14, 87, 178, 88, 95]
Outer_Lip = [61, 185, 40, 39, 37, 0, 267, 269, 270, 409, 291, 375, 321, 405, 314, 17, 84, 181, 91, 146]
Face_Boundary = [10, 338, 297, 332, 284, 251, 389, 356, 454, 323, 361, 288, 397, 365, 379, 378, 400, 377, 152, 148, 176, 149, 150, 136, 172, 58, 132, 93, 234, 127, 162, 21, 54, 103, 67, 109]
Nose = [168,6,197,195,5,4,1]
all_landmark_coords = Left_eyebrow + Right_eyebrow + Left_eye + Right_eye + Inner_Lip + Outer_Lip + Face_Boundary + Nose
landmark_coords_dict = dict(zip(all_landmark_coords,range(len(all_landmark_coords))))
# read the template image and detect landmarks
template_img = cv2.imread(template_path)
template_img = cv2.resize(template_img, (128, 128))
landmark_array = detect_landmark(template_img,(128, 128))
# scale the template image
min_val = landmark_array.min(axis=0)
max_val = landmark_array.max(axis=0)
norm_landmark_array = (landmark_array - min_val) / (max_val - min_val)
scaled_landmark_arr = norm_landmark_array * 80 + np.array([25, 25])
warp_points = get_parts(scaled_landmark_arr)
files = os.listdir(input_dir)
video_data_dict = {}
if not os.path.exists(output_dir):
os.mkdir(output_dir)
# iterate overfiles and align faces
for file in tqdm(files):
if file in corrupt_file_list:
continue
video_path = os.path.join(input_dir, file)
file_name = os.path.join(output_dir, file.replace(input_video_type, "pt"))
if not overwrite and os.path.exists(file_name):
continue
try:
video_frames = load_video_frames(video_path)
landmark_array = detect_landmark(video_frames[0],video_frames[0].shape[:2])
warp_points_2 = get_parts(landmark_array)
tform = tf.estimate_transform(
"similarity", warp_points, warp_points_2
) # find the transformation matrix
warped_frame_list = []
landmark_list = []
landmark_figure_list = []
for frame in video_frames:
color_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # turn images into RGB
warped_frame = (
tf.warp(color_frame, tform, output_shape=(128, 128)) * 255
) # warp the frame according to transformation
warped_frame = warped_frame.astype("uint8") # use uint8 to reduce file size
warped_frame_list.append(warped_frame)
temp_landmark = detect_landmark(warped_frame,(128,128))
if temp_landmark is not None:
landmark = temp_landmark.copy()
else:
landmark = np.full((135,2),np.nan) # if face not detected fill with nan
if not no_landmarks:
landmark_list.append(landmark)
landmark_array = np.array(landmark_list)
for i in range(landmark_array.shape[-1]):
for j in range(0,landmark_array.shape[1]):
ind = np.arange(landmark_array.shape[0])
good = np.where(~np.isnan(landmark_array[:,j,i]))[0]
f = interpolate.interp1d(ind[good],landmark_array[:,j,i][good]) # interpolate nan values
landmark_array[:,j,i] = f(ind)
landmark_array[:,j,i] = gaussian_filter1d(landmark_array[:,j,i], sigma=0.1)# filter each coordinate to reduce shaking
for i in range(landmark_array.shape[0]): # filter each coordinate
landmark_figure = draw_face(landmark_array[i].astype(int), size=(128, 128, 3))#, image = warped_frame_list[i]) # uncomment to draw landmarks on images
landmark_figure = cv2.cvtColor(landmark_figure, cv2.COLOR_BGR2RGB)
landmark_figure_list.append(landmark_figure)
warped_frame_array = np.array(warped_frame_list)
video_data_dict = {}
video_data_dict["video"] = warped_frame_array
if not no_landmarks:
landmark_array = np.array(landmark_list)
landmark_figure_array = np.array(landmark_figure_list)
video_data_dict["landmark"] = landmark_array
video_data_dict["landmark_figure"] = landmark_figure_array
torch.save(
video_data_dict,
file_name,
)
if test_video_output:
write_video(file_name,landmark_figure_array)
except Exception as e:
print(file) # Print file name and exception as extra caution
print(e)
with open(exception_file, "a") as output_file:
output_file.writelines(file + "\t" + str(e) + "\n")