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
| {
'camera_model': 'OPENCV', # camera model (currently only OpenCV is supported)
'height': 384, # height of the images
'width': 384, # width of the images
'has_mono_prior': true, # use monocular cues or not
'pairs': 'pairs.txt', # pairs file used for multi-view photometric consistency loss
'worldtogt': [
[1, 0, 0, 0], # world to gt transformation (useful for evauation)
[0, 1, 0, 0],
[0, 0, 1, 0],
[0, 0, 0, 1],
],
'scene_box': {
'aabb': [
[-1, -1, -1], # aabb for the bbox
[1, 1, 1],
],
'near': 0.5, # near plane for each image
'far': 4.5, # far plane for each image
'radius': 1.0, # radius of ROI region in scene
'collider_type': 'near_far',
# collider_type can be "near_far", "box", "sphere",
# it indicates how do we determine the near and far for each ray
# 1. near_far means we use the same near and far value for each ray
# 2. box means we compute the intersection with the bounding box
# 3. sphere means we compute the intersection with the sphere
},
'frames': [ # this contains information for each image
{
# note that all paths are relateive path
# path of rgb image
'rgb_path': '000000_rgb.png',
# camera to world transform
'camtoworld':
[
[
0.9702627062797546,
-0.014742869883775711,
-0.2416049987077713,
0.6601868867874146,
],
[
0.007479910273104906,
0.9994929432868958,
-0.03095100075006485,
0.07803472131490707,
],
[
0.2419387847185135,
0.028223417699337006,
0.9698809385299683,
-2.6397712230682373,
],
[0.0, 0.0, 0.0, 1.0],
],
# intrinsic of current image
'intrinsics':
[
[925.5457763671875, -7.8512319305446e-05, 199.4256591796875, 0.0],
[0.0, 922.6160278320312, 198.10269165039062, 0.0],
[0.0, 0.0, 1.0, 0.0],
[0.0, 0.0, 0.0, 1.0],
],
# path of monocular depth prior
'mono_depth_path': '000000_depth.npy',
# path of monocular normal prior
'mono_normal_path': '000000_normal.npy',
},
...,
],
}
|
