/
gen_npz.py
206 lines (176 loc) · 6.32 KB
/
gen_npz.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
"""
GTA-IM Dataset
"""
import glob
import os
import pickle
import numba
import numpy as np
@numba.jit(nopython=True, nogil=True)
def rot_axis(angle, axis):
cg = np.cos(angle)
sg = np.sin(angle)
if axis == 0: # X
v = [0, 4, 5, 7, 8]
elif axis == 1: # Y
v = [4, 0, 6, 2, 8]
else: # Z
v = [8, 0, 1, 3, 4]
RX = np.zeros(9, dtype=numba.float64)
RX[v[0]] = 1.0
RX[v[1]] = cg
RX[v[2]] = -sg
RX[v[3]] = sg
RX[v[4]] = cg
return RX.reshape(3, 3)
@numba.jit(nopython=True, nogil=True)
def rotate(vector, angle, inverse=False):
"""
Rotation of x, y, z axis
Forward rotate order: Z, Y, X
Inverse rotate order: X^T, Y^T,Z^T
Input:
vector: vector in 3D coordinates
angle: rotation along X, Y, Z (raw data from GTA)
Output:
out: rotated vector
"""
gamma, beta, alpha = angle[0], angle[1], angle[2]
# Rotation matrices around the X (gamma), Y (beta), and Z (alpha) axis
RX = rot_axis(gamma, 0)
RY = rot_axis(beta, 1)
RZ = rot_axis(alpha, 2)
# Composed rotation matrix with (RX, RY, RZ)
if inverse:
return np.dot(np.dot(np.dot(RX.T, RY.T), RZ.T), vector)
else:
return np.dot(np.dot(np.dot(RZ, RY), RX), vector)
def angle2rot(rotation, inverse=False):
return rotate(np.eye(3), rotation, inverse=inverse)
class Pose:
def __init__(self, position, rotation):
# relative position to the 1st frame: (X, Y, Z)
# relative rotation to the previous frame: (r_x, r_y, r_z)
self.position = position
self.rotation = angle2rot(rotation)
magic_rot = angle2rot(np.array([np.pi / 2, 0, 0]), inverse=True)
self.rotation = self.rotation.dot(magic_rot)
def get_focal_length(cam_near_clip, cam_field_of_view):
near_clip_height = (
2 * cam_near_clip * np.tan(cam_field_of_view / 2.0 * (np.pi / 180.0))
)
# camera focal length
return 1080.0 / near_clip_height * cam_near_clip
def get_cam_extr(cam_pos, cam_rot):
cam_pos = np.array(cam_pos)
cam_rot = np.array(cam_rot)
pose = Pose(cam_pos, cam_rot / 180.0 * np.pi)
cam_extr = np.eye(4)
cam_extr[:3, :3] = pose.rotation
cam_extr[:3, -1] = pose.position
return cam_extr
if __name__ == '__main__':
rec_inds = glob.glob('2020*')
for data_path in rec_inds:
if '.zip' in data_path:
continue
print(data_path)
data_path += '/'
info_path = data_path + 'realtimeinfo.gz'
info = pickle.load(open(info_path, 'rb'))['frames']
new_info = []
joints_2d_seq = []
joints_3d_cam_seq = []
joints_3d_world_seq = []
world2cam_trans = []
intrinsics = []
count = 0
for i in range(len(info)):
infot = info[i]
# Change the image names
prefix = data_path + str(infot['time'])
if os.path.exists(prefix + '_final.jpg') and os.path.exists(
prefix + '_depth.png'
):
os.rename(
prefix + '_final.jpg',
data_path + '{:05d}'.format(count) + '.jpg',
)
os.rename(
prefix + '_depth.png',
data_path + '{:05d}'.format(count) + '.png',
)
os.rename(
prefix + '_id.png',
data_path + '{:05d}'.format(count) + '_id.png',
)
count = count + 1
# 3d keypoints
keypoint = [
infot['head'],
infot['neck'],
infot['right_clavicle'],
infot['right_shoulder'],
infot['right_elbow'],
infot['right_wrist'],
infot['left_clavicle'],
infot['left_shoulder'],
infot['left_elbow'],
infot['left_wrist'],
infot['spine0'],
infot['spine1'],
infot['spine2'],
infot['spine3'],
infot['spine4'],
infot['right_hip'],
infot['right_knee'],
infot['right_ankle'],
infot['left_hip'],
infot['left_knee'],
infot['left_ankle'],
]
# camera parameters
cam_near_clip = infot['cam_near_clip']
cam_field_of_view = infot['cam_field_of_view']
focal_length = get_focal_length(
cam_near_clip, cam_field_of_view
)
intrinsic = np.asarray(
[
[focal_length, 0, 960.0],
[0, focal_length, 540.0],
[0, 0, 1],
]
)
cam_extr_ref = get_cam_extr(infot['cam_pos'], infot['cam_rot'])
joints = np.asarray(keypoint)
jn = joints.shape[0]
joints_world = np.concatenate(
[joints, np.ones((jn, 1))], axis=-1
)
joints_cam = joints_world.dot(np.linalg.inv(cam_extr_ref.T))[
:, :3
]
joints_2d = np.matmul(intrinsic, joints_cam.T)
joints_2d = (
joints_2d[0] / joints_2d[2],
joints_2d[1] / joints_2d[2],
)
gta_pose_2d = np.asarray(joints_2d).T.reshape(jn, 2)
joints_cam = joints_cam.reshape(jn, 3)
joints_2d_seq.append(np.asarray(joints_2d).T)
joints_3d_cam_seq.append(joints_cam)
joints_3d_world_seq.append(joints)
world2cam_trans.append(np.linalg.inv(cam_extr_ref.T))
intrinsics.append(intrinsic)
new_info.append(infot)
np.savez(
data_path + 'info_frames.npz',
joints_2d=np.asarray(joints_2d_seq),
joints_3d_cam=np.asarray(joints_3d_cam_seq),
joints_3d_world=np.asarray(joints_3d_world_seq),
world2cam_trans=np.asarray(world2cam_trans),
intrinsics=np.asarray(intrinsics),
)
fn = open(data_path + 'info_frames.pickle', 'wb')
pickle.dump(new_info, fn)