-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgenerate_flip_information.py
164 lines (109 loc) · 5.02 KB
/
generate_flip_information.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
from PIL import Image
import os
import shutil
import numpy as np
import math
import torch
import json
import re
dir_path = 'data/blender/ficus_test'
new_dir_path = dir_path + "_flip"
shutil.copytree(dir_path, new_dir_path)
new_dir_path = new_dir_path + '/train'
for filename in os.listdir(new_dir_path):
print("filename: ", filename, "flipped")
if filename.endswith('.png'):
img = Image.open(os.path.join(new_dir_path, filename))
img_flipped = img.transpose(Image.FLIP_LEFT_RIGHT)
base, ext = os.path.splitext(filename)
new_filename = f"{base}_flip{ext}"
img_flipped.save(os.path.join(new_dir_path, new_filename))
print(f"All images have been flipped and saved.")
def Find_Optimal_Sphere(X_cordi,Y_cordi,Z_cordi):
A = np.zeros((len(X_cordi),4))
A[:,0] = X_cordi*2
A[:,1] = Y_cordi*2
A[:,2] = Z_cordi*2
A[:,3] = 1
f = torch.zeros((len(X_cordi),1))
f[:,0] = (X_cordi*X_cordi) + (Y_cordi*Y_cordi) + (Z_cordi*Z_cordi)
C, residules, rank, singval = np.linalg.lstsq(A,f)
radius2 = (C[0]*C[0])+(C[1]*C[1])+(C[2]*C[2])+C[3]
radius = math.sqrt(radius2)
return radius, C[0], C[1], C[2]
def Symmetrically_Transforming(X_cordi,Y_cordi,Z_cordi, A,B,C,D):
X_flipped_cordi = []
Y_flipped_cordi = []
Z_flipped_cordi = []
for i in range(len(X_cordi)):
X_flipped_cordi.append(X_cordi[i] + 2*A*(-(A*X_cordi[i] + B*Y_cordi[i] + C*Z_cordi[i] + D)/(A*A+B*B+C*C)))
Y_flipped_cordi.append(Y_cordi[i] + 2*B*(-(A*X_cordi[i] + B*Y_cordi[i] + C*Z_cordi[i] + D)/(A*A+B*B+C*C)))
Z_flipped_cordi.append(Z_cordi[i] + 2*C*(-(A*X_cordi[i] + B*Y_cordi[i] + C*Z_cordi[i] + D)/(A*A+B*B+C*C)))
X_flipped_cordi = torch.FloatTensor(X_flipped_cordi)
Y_flipped_cordi = torch.FloatTensor(Y_flipped_cordi)
Z_flipped_cordi = torch.FloatTensor(Z_flipped_cordi)
return X_flipped_cordi, Y_flipped_cordi, Z_flipped_cordi
def Projection2Sphere(x,y,z, x0,y0,z0,r):
a = x*x + y*y + z*z
b = x*x0 + y*y0 + z*z0
c = x0*x0 + y0*y0 + z0*z0 -r*r
alpha1 = (-b + math.sqrt(b*b - 4*a*c) )/(2*a)
alpha2 = (-b - math.sqrt(b*b - 4*a*c) )/(2*a)
if abs(alpha1-1) > abs(alpha2-1):
alpha = alpha2
else:
alpha = alpha1
return alpha*x, alpha*y, alpha*z
up = torch.FloatTensor([0,0,1])
def Get_Transform_Matrics(x,y,z, x0,y0,z0, up):
c = torch.stack([x,y,z])
c = c.squeeze()
at = torch.FloatTensor([x0,y0,z0])
at = at.squeeze()
torch0 = torch.FloatTensor([0.])
temp = torch.FloatTensor([0,0,0,1])
z_axis = (c-at) / torch.norm(c-at)
x_axis = torch.cross(up, z_axis) / torch.norm(torch.cross(up, z_axis))
y_axis = torch.cross(z_axis,x_axis)
z = torch.cat([z_axis, torch0])
x = torch.cat([x_axis, torch0])
y = torch.cat([y_axis, torch0])
rotation_mat = torch.stack([x, y, z, temp], dim=-1)
translation_mat = torch.eye(4)
translation_mat[:3,3] = c
transform_mat = torch.matmul(translation_mat, rotation_mat)
return transform_mat
up = torch.FloatTensor([0,0,1])
def Find_Optimal_CameraPose (X_cordi, Y_cordi, Z_cordi):
flipped_camera_pose_set = []
r, x0, y0, z0 = Find_Optimal_Sphere(X_cordi,Y_cordi,Z_cordi)
X_flipped_cordi, Y_flipped_cordi, Z_flipped_cordi = Symmetrically_Transforming(X_cordi,Y_cordi,Z_cordi, 1,0,0,0)
for i in range(len(X_cordi)):
X_optim, Y_optim, Z_optim = Projection2Sphere(X_flipped_cordi[i], Y_flipped_cordi[i], Z_flipped_cordi[i], x0,y0,z0,r)
transform_mat = Get_Transform_Matrics(X_optim.float(), Y_optim.float(), Z_optim.float(), x0,y0,z0, up)
flipped_camera_pose_set.append(transform_mat)
return flipped_camera_pose_set
def make_flipped_json(file_path, output_path):
with open(file_path, "r") as file:
data = json.load(file)
new_frames = []
X_cordi = [frame["transform_matrix"][0][3] for frame in data["frames"]]
Y_cordi = [frame["transform_matrix"][1][3] for frame in data["frames"]]
Z_cordi = [frame["transform_matrix"][2][3] for frame in data["frames"]]
result = Find_Optimal_CameraPose(torch.FloatTensor(X_cordi), torch.FloatTensor(Y_cordi), torch.FloatTensor(Z_cordi))
result = [item.tolist() for item in result]
for i, frame in enumerate(data["frames"]):
new_frame = frame.copy()
new_frame["file_path"] = new_frame["file_path"] + "_flip"
new_frame["transform_matrix"] = result[i]
new_frames.append(new_frame)
data["frames"].extend(new_frames)
data["frames"].sort(key=lambda frame: (int(re.search(r'\d+', frame["file_path"]).group()), "_flip" in frame["file_path"]))
with open(output_path, "w") as file:
json.dump(data, file, indent=4)
base_path = dir_path + '/'
new_dir_path_j = new_dir_path + '/'
input_file_path = os.path.join(base_path, "transforms_train.json")
output_file_path = os.path.join(new_dir_path_j, "flipped_transforms_train.json")
make_flipped_json(input_file_path, output_file_path)
os.rename(output_file_path, input_file_path)