-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
308 lines (251 loc) · 9.43 KB
/
utils.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
from __future__ import division, print_function
import matplotlib.pyplot as plt
from glob import glob
import subprocess
from webvtt import WebVTT
import argparse
import os
import pickle as pkl
from tqdm import tqdm
from sklearn.decomposition import PCA
import numpy as np
from shutil import rmtree
import soundfile as sf
import pyworld as pw
import scipy.io.wavfile as wav
from python_speech_features import logfbank
import cv2
import sys
import dlib
from skimage import io
from imutils import face_utils
import imutils
from keras.models import Sequential
from keras.layers import Dense, LSTM, Dropout, Embedding, Lambda, TimeDistributed
import keras.backend as K
from keras.preprocessing.sequence import pad_sequences
from keras.models import load_model
import keras
from sklearn.preprocessing import MinMaxScaler
import gc
import time
predictor_path = 'shape_predictor_68_face_landmarks.dat'
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor(predictor_path)
#####################################################################
n_batch = 1 # Conservative guess on the batchsize
time_delay = 10
length = 60 # More than this and the LSTM becomes retard
#####################################################################
# trim_time = 3.0
# delay_in_seconds = 0 # 0.200
# time_delay = int(np.ceil(delay_in_seconds * 100)) # 150 ms and 200 fps
# limit_up = int(trim_time*100) # 500
# train_val_ratio = 0.8
# batchSize = 100
def get_facial_landmarks(filename):
image = io.imread(filename);
# detect face(s)
dets = detector(image, 1);
shape = np.empty([1,1])
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = predictor(image, d);
shape = face_utils.shape_to_np(shape);
return shape
def getTilt(keypoints_mn):
# Remove in plane rotation using the eyes
eyes_kp = np.array(keypoints_mn[36:47])
x = eyes_kp[:, 0]
y = -1*eyes_kp[:, 1]
# print('X:', x)
# print('Y:', y)
m = np.polyfit(x, y, 1)
tilt = np.degrees(np.arctan(m[0]))
return tilt
def drawLips(keypoints, new_img, c = (255, 255, 255), th = 1, show = False):
keypoints = np.float32(keypoints)
for i in range(48, 59):
cv2.line(new_img, tuple(keypoints[i]), tuple(keypoints[i+1]), color=c, thickness=th)
cv2.line(new_img, tuple(keypoints[48]), tuple(keypoints[59]), color=c, thickness=th)
cv2.line(new_img, tuple(keypoints[48]), tuple(keypoints[60]), color=c, thickness=th)
cv2.line(new_img, tuple(keypoints[54]), tuple(keypoints[64]), color=c, thickness=th)
cv2.line(new_img, tuple(keypoints[67]), tuple(keypoints[60]), color=c, thickness=th)
for i in range(60, 67):
cv2.line(new_img, tuple(keypoints[i]), tuple(keypoints[i+1]), color=c, thickness=th)
if (show == True):
cv2.imshow('lol', new_img)
cv2.waitKey(10000)
def getKeypointFeatures(keypoints):
# Mean Normalize the keypoints wrt the center of the mouth
# Leads to face position invariancy
mouth_kp_mean = np.average(keypoints[48:67], 0)
keypoints_mn = keypoints - mouth_kp_mean
# Remove tilt
x_dash = keypoints_mn[:, 0]
y_dash = keypoints_mn[:, 1]
theta = np.deg2rad(getTilt(keypoints_mn))
c = np.cos(theta); s = np.sin(theta)
x = x_dash*c - y_dash*s # x = x'cos(theta)-y'sin(theta)
y = x_dash*s + y_dash*c # y = x'sin(theta)+y'cos(theta)
keypoints_tilt = np.hstack((x.reshape((-1,1)), y.reshape((-1,1))))
# Normalize
N = np.linalg.norm(keypoints_tilt, 2)
return [keypoints_tilt/N, N, theta, mouth_kp_mean]
def getOriginalKeypoints(kp_features_mouth, N, tilt, mean):
# Denormalize the points
kp_dn = N * kp_features_mouth
# Add the tilt
x, y = kp_dn[:, 0], kp_dn[:, 1]
c, s = np.cos(tilt), np.sin(tilt)
x_dash, y_dash = x*c + y*s, -x*s + y*c
kp_tilt = np.hstack((x_dash.reshape((-1,1)), y_dash.reshape((-1,1))))
# Shift to the mean
kp = kp_tilt + mean
return kp
def get_sec(time_str):
h, m, s = time_str.split(':')
return int(h) * 3600 + int(m) * 60 + float(s)
####################################################################################
# The model
def LSTM_lipsync(in_shape = (n_batch, length, 26), out_shape = (length, 8)):
# model = Sequential()
# model.add(LSTM(256, batch_input_shape=in_shape, return_sequences=True)) #, stateful=True))
# model.add(TimeDistributed(Dense(out_shape[1])))
# model.compile(loss='mean_squared_error', optimizer='adam')
# print(model.summary())
# return model
model = Sequential()
model.add(LSTM(8, input_shape=(length, 26)))
model.compile(loss='mean_squared_error', optimizer='adam')
print(model.summary())
return model
def batchify(X, n_batch): # X is a 3D array
X = np.array(X)
n = X.shape[0] % n_batch
# print('n:', n, 'sub:', n_batch-n)
Z = np.zeros((n_batch-n, length, X.shape[2]))
X = np.vstack((X, Z))
return X
# Get the data into proper format
# i.e [samples, timesteps, features]
# where num of samples should be integral multiple of n_batches
def getData(audio_kp, video_kp, pca, nTrainingVideo):
# Total number of elements in each list
# print('len(audio):', len(audio_kp), 'len(video):', len(video_kp))
X, y = [], [] # Create the empty lists
# Get the common keys
keys_audio = audio_kp.keys()
keys_video = video_kp.keys()
keys = sorted(list(set(keys_audio).intersection(set(keys_video))))
# print('Length of common keys:', len(keys), 'First common key:', keys[0])
for key in tqdm(keys[0:nTrainingVideo]):
audio = audio_kp[key]
video = video_kp[key]
# Get the lesser size of the two matrices
n_lesser = len(audio) if (len(audio) < len(video)) else len(video)
# Need to get smaller timesteps from this huge data
segregateTimesteps = int(np.floor((n_lesser-time_delay)/length))
# print('seg:', segregateTimesteps, 'n_lesser:', n_lesser, 'length:', length)
# Stuff chunks of this juicy data into the x and y
for i in range(segregateTimesteps):
X.append(audio[i*length+time_delay: (i+1)*length+time_delay])
y.append(video[i*length: (i+1)*length])
# # normalize the dataset
# scalerX = MinMaxScaler(feature_range=(0, 1))
# scalerY = MinMaxScaler(feature_range=(0, 1))
# X = np.array(X)
# X = X.reshape(X.shape[0]*X.shape[1], X.shape[2])
# y = np.array(y)
# y = y.reshape(y.shape[0]*y.shape[1], y.shape[2])
# # print('Shape of X:', X.shape)
# X = scalerX.fit_transform(X)
# y = scalerY.fit_transform(y)
# X = X.reshape(int(X.shape[0]/length), length, X.shape[1])
# y = y.reshape(int(y.shape[0]/length), length, y.shape[1])
X = batchify(X, n_batch)
y = batchify(y, n_batch)
n = X.shape[0]
val_flag = False
if n >= 5*n_batch: # this is where we have a validation set
split = int(n*0.8)
split = int(np.ceil(split/n_batch)*n_batch)
val_flag = True
else: # no validation set
split = n
train_X = X[0:split]
val_X = X[split:]
train_y = y[0:split]
val_y = y[split:]
return train_X, train_y, val_X, val_y, val_flag
def preparekpForPrediction(audio_kp):
# Need to get smaller timesteps from this huge data
segregateTimesteps = int(np.floor((audio_kp.shape[0]-time_delay)/length))
# Stuff chunks of this juicy data into the X
X = []
for i in range(segregateTimesteps):
X.append(audio_kp[i*length+time_delay: (i+1)*length+time_delay, :])
X = np.array(X)
X = batchify(X, n_batch)
return X
def audioToPrediction(filename):
# Get audio features
(rate, sig) = wav.read(filename)
audio_kp = logfbank(sig,rate)
originalNumofPts = audio_kp.shape[0]
return preparekpForPrediction(audio_kp), originalNumofPts
def subsample(y, fps_from = 100.0, fps_to = 29.97):
factor = int(np.ceil(fps_from/fps_to))
# Subsample the points
new_y = np.zeros((int(y.shape[0]/factor), 20, 2)) #(timesteps, 20) = (500, 20x2)
for idx in range(new_y.shape[0]):
if not (idx*factor > y.shape[0]-1):
# Get into (x, y) format
new_y[idx, :, 0] = y[idx*factor, 0:20]
new_y[idx, :, 1] = y[idx*factor, 20:]
else:
break
# print('Subsampled y:', new_y.shape)
new_y = [np.array(each) for each in new_y.tolist()]
# print(len(new_y))
return new_y
# Get the data into proper format
# i.e [samples, timesteps, features]
# where num of samples should be integral multiple of n_batches
def getDataNormalized(audio_kp, video_kp, pca, nTrainingVideo):
# Total number of elements in each list
# print('len(audio):', len(audio_kp), 'len(video):', len(video_kp))
X, y = np.zeros((1, 26)), np.zeros((1, 8)) # Create the empty lists
# Get the common keys
keys_audio = audio_kp.keys()
keys_video = video_kp.keys()
keys = sorted(list(set(keys_audio).intersection(set(keys_video))))
# print('Length of common keys:', len(keys), 'First common key:', keys[0])
for key in tqdm(keys[0:nTrainingVideo]):
audio = audio_kp[key]
video = video_kp[key]
# Get the lesser size of the two matrices
n_lesser = len(audio) if (len(audio) < len(video)) else len(video)
# print('audio shape:', audio.shape)
X = np.vstack((X, audio[0+time_delay:n_lesser+time_delay]))
y = np.vstack((y, video[0:n_lesser]))
# normalize the dataset
scalerX = MinMaxScaler(feature_range=(0, 1))
scalerY = MinMaxScaler(feature_range=(0, 1))
X = np.array(X)
y = np.array(y)
# print('Shape of X:', X.shape)
X = scalerX.fit_transform(X)
y = scalerY.fit_transform(y)
split = 13000
train_X = X[0:split]
val_X = X[split:]
train_y = y[0:split]
val_y = y[split:]
numberOfSamples_train = 100;
numberOfSamples_val = 10;
train_X = train_X[0:numberOfSamples_train*length].reshape((-1, length, 26))
train_y = train_y[0:numberOfSamples_train*length].reshape((-1, length, 8))
val_X = val_X[0:numberOfSamples_val*length].reshape((-1, length, 26))
val_y = val_y[0:numberOfSamples_val*length].reshape((-1, length, 8))
return train_X, train_y, val_X, val_y, scalerX, scalerY