forked from CMU-TBD/Group_based_navigation_v1
-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_generation.py
168 lines (143 loc) · 6.29 KB
/
data_generation.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
import cv2
import numpy as np
from message import Message
from data_loader import DataLoader
from grouping import Grouping
from group_shape_generation import GroupShapeGeneration
from img_process import ProcessImage, DrawGroupShape
class DataGeneration(object):
def __init__(self, history, offset, train_set, test_set):
self.fps = 10
self.history = history
self.og_history = history
#self.history += 1
self.history += 8
self.offset = offset
self.history *= offset
self.msg_group = self._get_msg_group()
self.train_idx = train_set
self.test_idx = test_set
self.train_prob = self._calculate_set_prob(self.train_idx)
self.test_prob = self._calculate_set_prob(self.test_idx)
self.group_labels, self.frame_labels = self._get_labels()
self.num_train_data = 0
self.num_test_data = 0
for e in self.train_idx:
self.num_train_data += len(self.group_labels[e])
for e in self.test_idx:
self.num_test_data += len(self.group_labels[e])
return
def _get_msg_group(self):
msg_group = []
dataset_list = ['eth', 'eth', 'ucy', 'ucy', 'ucy']
dataset_idx_list = [0, 1, 0, 1, 2]
#dataset_list = ['eth']
#dataset_idx_list = [0]
for i in range(len(dataset_list)):
dataset = dataset_list[i]
dataset_idx = dataset_idx_list[i]
msg = Message()
data = DataLoader(dataset, dataset_idx, self.fps)
msg = data.update_message(msg)
gp = Grouping(msg, self.history)
msg = gp.update_message(msg)
msg_group.append(msg)
return msg_group
def _calculate_set_prob(self, set_idx):
set_prob = []
for i in set_idx:
msg = self.msg_group[i]
gp_labels = msg.video_labels_matrix
set_prob.append(len(self._get_unique_labels(gp_labels)))
set_prob = np.array(set_prob)
set_prob = set_prob / np.sum(set_prob)
return set_prob
def _get_labels(self):
# group_labels: all groups ids that exist longer than history
# frame_labels: valid frames for the groups to sample from (accounting for history)
group_labels = []
frame_labels = []
for msg in self.msg_group:
msg_group_labels = []
msg_frame_labels = []
tmp_frame_labels = []
labels = msg.video_labels_matrix
max_label = np.max(self._get_unique_labels(labels))
for i in range(max_label + 1):
tmp_frame_labels.append([])
for i, sub_list in enumerate(labels):
for elem in sub_list:
tmp_frame_labels[elem].append(i)
for i, sub_list in enumerate(tmp_frame_labels):
sub_list = np.unique(sub_list)
if not(len(sub_list) < self.history):
msg_group_labels.append(i)
msg_frame_labels.append(sub_list[(self.history - 1):None])
group_labels.append(msg_group_labels)
frame_labels.append(msg_frame_labels)
return group_labels, frame_labels
def _get_unique_labels(self, labels):
all_labels = []
for sub_list in labels:
all_labels += sub_list
return np.unique(all_labels)
def generate_sample(self, from_train=True, debug=False):
if from_train:
idx = np.random.choice(self.train_idx, p=self.train_prob)
else:
idx = np.random.choice(self.test_idx, p=self.test_prob)
msg = self.msg_group[idx]
shape_gen_class = GroupShapeGeneration(msg)
group_pool = self.group_labels[idx]
#print(len(group_pool))
frame_pool = self.frame_labels[idx]
if (len(group_pool) == 0):
raise Exception('No valid groups exist!')
group_idx = np.random.choice(range(len(group_pool)))
group = group_pool[group_idx]
frame = np.random.choice(frame_pool[group_idx])
img_seq = self._generate_img_sequence(shape_gen_class, msg, group, frame, debug, from_train)
#return np.array(img_seq[:-1]), np.array(img_seq[-1])
return np.array(img_seq[:self.og_history]), np.array(img_seq[self.og_history:])
def generate_cases_all_groups(self, case_num):
msg = self.msg_group[case_num]
shape_gen_class = GroupShapeGeneration(msg)
group_pool = self.group_labels[case_num]
frame_pool = self.frame_labels[case_num]
num_groups = len(group_pool)
input_cases = []
output_cases = []
for i in range(num_groups):
group = group_pool[i]
frame = np.random.choice(frame_pool[i])
img_seq = self._generate_img_sequence(shape_gen_class, msg, group, frame, False, True)
input_cases.append(np.array(img_seq[:self.og_history]))
output_cases.append(np.array(img_seq[self.og_history:]))
return input_cases, output_cases
def _generate_img_sequence(self, shape_gen_class, msg, group, frame, debug, from_train):
norm_ang = False
vertice_sequence = []
all_group_info = []
for i in range(frame - self.history + 1, frame + 1, self.offset):
vertices, group_info = shape_gen_class.generate_group_shape(i, group)
vertice_sequence.append(vertices)
all_group_info.append(group_info)
dgs = DrawGroupShape(msg)
dgs.set_center(vertice_sequence[:self.og_history])
if norm_ang:
# This is still bugged
velocities = all_group_info[self.og_history - 1][1]
avg_vel = dgs.coordinate_transform(np.mean(np.array(velocities), axis=0))
aug_ang = np.arctan2(avg_vel[1], avg_vel[0]) / np.pi * 180
else:
aug_ang = None
dgs.set_aug(angle=aug_ang)
img_sequence = []
for i, v in enumerate(vertice_sequence):
canvas = np.zeros((msg.frame_height, msg.frame_width, 3), dtype=np.uint8)
img = dgs.draw_group_shape(v, canvas, center=True, aug=from_train)
img_sequence.append(img)
pimg = ProcessImage(msg, img_sequence[:-1])
for i, img in enumerate(img_sequence):
img_sequence[i] = pimg.process_image(img, debug)
return img_sequence