-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathdataset_README.txt
124 lines (92 loc) · 5.2 KB
/
dataset_README.txt
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
Dataset file structure:
data/
hg/
scene1_hg.json
scene2_hg.json
scene3_hg.json
mask/
scene1/
p1c1_mask.png (4k)
p1c1_mask_1080.png (1080p)
...
scene2/
...
scene3/
...
obj/
scene1_annotations.csv
scene2_annotations.csv
scene3_annotations.csv
spl_obj/
scene1_splobj.json
scene2_splobj.json
scene3_splobj.json
track/
scene1_tracklets.json
scene2_tracklets.json
scene3_tracklets.json
ts/
scene1_ts.csv
scene2_ts.csv
scene3_ts.csv
scene1_ts_orig.csv
scene2_ts_orig.csv
scene3_ts_orig.csv
video/
scene1/
p1c1.mp4
p1c2.mp4
...
scene2/
...
scene3/
...
cache/
This directory is empty but is designated as the default location for saving image files to disk
File Formats:
sceneX_hg.json
A dictionary of camera correspondences, keyed by camera name (e.g. "p1c1"). Each object is itself a dictionary with the following keys:
corr_pts - [n,2] array of correspondence points labeled in imagery
space_pts - [n,2] array of correspondence points labeled in overhead roadway imagery (space). All points lie on z=0 plane by assumption.
P - [4,3] space-> image projection matrix
H - [3,3] image plane -> ground plane homography matrix
H_inv - [3,3] ground plane -> image plane homography matrix (3 of 4 columns are identical to P)
curve - 3 parameters fitting a second order polynomial to the curvature of the roadway, in imagery (see homography.py for usage)
vps - list of 3 pairs of points for (x,y) pixel coordinates for the longitudinal, lateral, and vertical vanishing points, respectively
sceneX_annotations.csv
Each file is sorted by increasing frame index. This is the required file format for scene.py
frame camera id x y l w h direction class gen
0 p1c1 0 372.043 4.141 15.924 5.722 4.48 1 sedan manual
0 p1c1 1 380.529 24.258 17.183 5.759 5.98 1 pickup manual
0 p1c1 2 361.166 43.482 15.075 5.479 5.58 1 midsize manual
0 p1c1 3 437.605 17.869 18.518 5.937 6.04 1 pickup manual
0 p1c1 46 464.146 103.813 35.711 8.57 12.02 -1 truck manual
0 p1c1 47 368.431 77.674 14.019 5.805 4.82 -1 midsize manual
0 p1c1 48 321.267 90.856 14.507 6.365 4.86 -1 midsize manual
...
direction - 1 for all EB (near side relative to cameras) vehicles and -1 for all WB vehicles
gen - "manual" if point was manually clicked, "interpolation" if point was interpolated between manual points and subsequently inspected without modification, or "spline" if object was produced by sampling the spline approximation of the object
camera - <pole#><camera#>
x,y,l,w,h - position and dimensions in feet
id - unique integer for each unique object
sceneX_splobj.json and sceneX_tracklets.json
For tracking evaluation, objects are stored rather than storing data indexed by time. This allows for easy reinterpolation of the target and predicted object set to the same discrete times. This is the required format for evaluate.py.
Each file contains a list of objects. Each object is a dictionary with the following keys:
id - int
class - str (from sedan, midsize, pickup, van, semi, truck (other))
l - float (length in feet)
w - float (width in feet)
h - float (height in feet)
direction - int (-1 or 1 for WB or EB)
x_position - 1D float array (x position in feet)
y_position - 1D float array (y position in feet)
timestamp - 1D float array (timestamp in s)
sceneX_ts.csv and sceneX_ts_orig.csv
First column is the frame index, all following columns are the corrected timestamp or original timestamp, respectively, in seconds, for each camera in the scene:
frame p1c1 p1c2 p1c3 p1c4 p1c5 p1c6 p2c1 p2c2 p2c3 p2c4 p2c5 p2c6 p3c1 p3c2 p3c3 p3c4 p3c5
0 1623877088.8 1623877088.8075 1623877088.8213 1623877088.8193 1623877088.8284 1623877088.7236 1623877088.7908 1623877088.7235 1623877088.7934 1623877088.7335 1623877088.8369 1623877088.7406 1623877089.3395 1623877089.4021 1623877089.6324 1623877089.6127 1623877090.7898
1 1623877088.8357 1623877088.8404 1623877088.8527 1623877088.8492 1623877088.8685 1623877088.7636 1623877088.8309 1623877088.7635 1623877088.8305 1623877088.7677 1623877088.8727 1623877088.7734 1623877089.3651 1623877089.435 1623877089.6667 1623877089.6527 1623877090.8241
2 1623877088.8642 1623877088.8818 1623877088.8942 1623877088.8806 1623877088.9113 1623877088.798 1623877088.8738 1623877088.7949 1623877088.862 1623877088.8006 1623877088.9041 1623877088.8034 1623877089.3965 1623877089.4664 1623877089.6982 1623877089.6827 1623877090.8627
3 1623877088.8957 1623877088.9133 1623877088.9299 1623877088.9121 1623877088.9442 1623877088.8309 1623877088.9067 1623877088.8278 1623877088.8934 1623877088.8335 1623877088.9456 1623877088.8349 1623877089.4279 1623877089.4979 1623877089.7296 1623877089.7127 1623877090.8927
4 1623877088.9357 1623877088.9433 1623877088.9599 1623877088.9535 1623877088.9771 1623877088.8752 1623877088.9382 1623877088.8693 1623877088.9348 1623877088.8763 1623877088.9756 1623877088.8763 1623877089.4694 1623877089.5379 1623877089.7595 1623877089.7427 1623877090.9227
...