-
Notifications
You must be signed in to change notification settings - Fork 0
/
feature_maps.py
200 lines (146 loc) · 6.63 KB
/
feature_maps.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
import numpy as np
import os
import glob
from utils import sorted_nicely, mkGaussian, compute_density_image, clean_eyedata
import cv2
from feat_map import Feat_map
class Feature_maps(object):
def __init__(self, dynDir=None, facemapDir=None):
self.dynDir = dynDir
self.facemapDir = facemapDir
self.all_fmaps = []
def release_fmaps(self):
self.all_fmaps = []
def load_feature_maps(self, video_name, vidHeight=None, vidWidth=None):
sts_saliency_maps = os.listdir(self.dynDir + video_name[:-4])
sorted_sts_saliency_maps = sorted_nicely(sts_saliency_maps)
self.num_sts = len(sorted_sts_saliency_maps)
face1_maps = glob.glob(self.facemapDir + video_name[:-4] + '/*_speaker.png')
sorted_face1_maps = sorted_nicely(face1_maps)
self.num_speak = len(sorted_face1_maps)
face2_maps = glob.glob(self.facemapDir + video_name[:-4] + '/*_nonspeaker.png')
sorted_face2_maps = sorted_nicely(face2_maps)
self.num_nspeak = len(sorted_face2_maps)
mu = np.array([vidWidth/2,vidHeight/2])
wy=vidHeight/12
wx=vidWidth/12
sigma = [wx, wy]
F = mkGaussian(mu, sigma, 0, vidWidth, vidHeight).T
center_bias_map = Feat_map(feat_map=F/np.sum(np.sum(F)), name='CB')
uniform_map = Feat_map(feat_map=np.ones(center_bias_map.shape)/np.prod(center_bias_map.shape), name='Uniform')
self.sts_names = sorted_sts_saliency_maps
self.face1_names = sorted_face1_maps
self.face2_names = sorted_face2_maps
self.cb = center_bias_map
self.uniform = uniform_map
self.video_name = video_name[:-4]
self.vidHeight = vidHeight
self.vidWidth = vidWidth
return sorted_sts_saliency_maps, sorted_face1_maps, sorted_face2_maps, center_bias_map, uniform_map
def read_current_maps(self, gaze, frame_num, compute_heatmap=False):
curr_sts = cv2.imread(self.dynDir + self.video_name + '/' + self.sts_names[frame_num], 0)
self.sts = Feat_map(feat_map=curr_sts/float(np.sum(curr_sts)), name='STS')
curr_sts = np.reshape(curr_sts,-1, order='F')
curr_face1 = cv2.imread(self.face1_names[frame_num], 0)
if np.sum(curr_face1) != 0:
self.speaker = Feat_map(feat_map=curr_face1/float(np.sum(curr_face1)), name='Speaker')
curr_face1 = np.reshape(curr_face1,-1, order='F')/np.max(curr_face1)
else:
self.speaker = Feat_map(feat_map=curr_face1, name='Speaker')
curr_face1 = np.reshape(curr_face1,-1, order='F')
curr_face2 = cv2.imread(self.face2_names[frame_num], 0)
if np.sum(curr_face2) != 0:
self.non_speaker = Feat_map(feat_map=curr_face2/float(np.sum(curr_face2)), name='non_Speaker')
curr_face2 = np.reshape(curr_face2,-1, order='F')/np.max(curr_face2)
else:
self.non_speaker = Feat_map(feat_map=curr_face2, name='non_Speaker')
curr_face2 = np.reshape(curr_face2,-1, order='F')
self.all_fmaps.append(self.cb)
self.all_fmaps.append(self.sts)
self.all_fmaps.append(self.speaker)
self.all_fmaps.append(self.non_speaker)
if compute_heatmap:
try:
w = self.vidWidth
h = self.vidHeight
curr_gaze = clean_eyedata(gaze[:,frame_num,:], w, h)
nObsTrain = int(np.floor(curr_gaze.shape[0]*1))
Eye_Position_Map_train = np.zeros([w,h])
curr_gaze_train_ind = curr_gaze[:nObsTrain,:].astype(int)
Eye_Position_Map_train = compute_density_image(curr_gaze_train_ind, [w,h])
self.original_eyeMap = Eye_Position_Map_train.copy()
except:
pass
def read_current_maps_v2(self, gaze, frame_num, train_perc, num_features, d_rate=None, downsample=True, ):
if not downsample:
d_rate = 100
w = self.vidWidth
h = self.vidHeight
nMaps = num_features #questo valore definisce quante FutureMaps voglio
curr_sts = cv2.imread(self.dynDir + self.video_name + '/' + self.sts_names[frame_num], 0)
wd = int(curr_sts.shape[1] * d_rate / 100)
hd = int(curr_sts.shape[0] * d_rate / 100)
dim = (wd, hd)
self.sts = Feat_map(feat_map=curr_sts/float(np.sum(curr_sts)), name='STS')
if downsample:
curr_sts = cv2.resize(curr_sts, dim, interpolation=cv2.INTER_AREA)
curr_sts = np.reshape(curr_sts,-1, order='F')
newShape = wd*hd #New Shape after downsampling
X = np.empty([newShape, nMaps])
curr_face1 = cv2.imread(self.face1_names[frame_num], 0)
if np.sum(curr_face1) != 0:
self.speaker = Feat_map(feat_map=curr_face1/float(np.sum(curr_face1)), name='Speaker')
if downsample:
curr_face1 = cv2.resize(curr_face1, dim, interpolation=cv2.INTER_AREA)
curr_face1 = np.reshape(curr_face1,-1, order='F')/np.max(curr_face1)
else:
self.speaker = Feat_map(feat_map=curr_face1, name='Speaker')
if downsample:
curr_face1 = cv2.resize(curr_face1, dim, interpolation=cv2.INTER_AREA)
curr_face1 = np.reshape(curr_face1,-1, order='F')
curr_face2 = cv2.imread(self.face2_names[frame_num], 0)
if np.sum(curr_face2) != 0:
self.non_speaker = Feat_map(feat_map=curr_face2/float(np.sum(curr_face2)), name='non_Speaker')
if downsample:
curr_face2 = cv2.resize(curr_face2, dim, interpolation=cv2.INTER_AREA)
curr_face2 = np.reshape(curr_face2,-1, order='F')/np.max(curr_face2)
else:
self.non_speaker = Feat_map(feat_map=curr_face2, name='non_Speaker')
if downsample:
curr_face2 = cv2.resize(curr_face2, dim, interpolation=cv2.INTER_AREA)
curr_face2 = np.reshape(curr_face2,-1, order='F')
self.all_fmaps.append(self.uniform)
self.all_fmaps.append(self.cb)
self.all_fmaps.append(self.sts)
self.all_fmaps.append(self.speaker)
self.all_fmaps.append(self.non_speaker)
if downsample:
cbValue = cv2.resize(self.cb.value, dim, interpolation=cv2.INTER_AREA)
unifValue = cv2.resize(self.uniform.value, dim, interpolation=cv2.INTER_AREA)
else:
cbValue = self.cb.value
unifValue = self.uniform.value
cb_reshapaed = np.reshape(cbValue,-1, order='F')
uniform_reshaped = np.reshape(unifValue,-1, order='F')
X[:,0] = curr_sts
X[:,1] = cb_reshapaed
X[:,2] = uniform_reshaped
X[:,3] = curr_face1
X[:,4] = curr_face2
curr_gaze = clean_eyedata(gaze[:,frame_num,:], w, h)
nObsTrain = int(np.floor(curr_gaze.shape[0]*train_perc))
Eye_Position_Map_train = np.zeros([w,h])
Eye_Position_Map_test = np.zeros([w,h])
curr_gaze_train_ind = curr_gaze[:nObsTrain,:].astype(int)
curr_gaze_test_ind = curr_gaze[nObsTrain:,:].astype(int)
Eye_Position_Map_train = compute_density_image(curr_gaze_train_ind, [w,h])
if train_perc < 1:
Eye_Position_Map_test = compute_density_image(curr_gaze_test_ind, [w,h])
else:
Eye_Position_Map_test = np.zeros([w,h])
self.X = X
self.original_eyeMap = Eye_Position_Map_train.copy()
if downsample:
Eye_Position_Map_train = cv2.resize(Eye_Position_Map_train, dim, interpolation=cv2.INTER_AREA)
self.eyeMap_train = Eye_Position_Map_train
return X, Eye_Position_Map_train, Eye_Position_Map_test