Chameleon Annotation and Meta data

Introduction

The annotations file is a csv format text file with a one-line entry for each tracked object visible in each output image. Objects are tracked by category and subcategory as specified in the simulation template used for the simulation run. If an object category is not being tracked there will be no entries for it in the annotations file.

Some of the entries may be linked to licensed features; if an entry is not present, check your license conditions.

Ids and Category

Name Description Type
idx Unique id for every entry in a dataset. Contiguity is not guaranteed. integer
unique_id Unique id for every instance of an object. This is used to train object 
trackers. Integer id can be any value, but must be unique to that object 
instance for every appearance of that object instance in the dataset. The id persists only within a simulation run and is not guaranteed to be the same on subsequent runs. This value is used as the mask id value in the mask output file if one is being output.
integer
parent_id Unique id of the parent object of this body part. zero if this is not a 
body part. The id persists only within a simulation run and is not guaranteed to be the same on subsequent runs.
integer
cat_id Globally unique category integer value starting at 1 
(ONE). 0 (ZERO) is RESERVED. Object categories are assigned when assets are created (outside the simulator).
integer
cat_text Unique text descriptor for each category integer value. string
subcat_id Locally unique sub category index starting at 1 (ONE). 0 (ZERO) is RESERVED. Id numbers are unique within the relevant category. integer
subcat_text Text descriptor for subcategory. Unique for each subcategory. string

Camera

Name Description Type
cam_x_pos World space x position of camera in metres from origin. float
cam_y_pos World space y position of camera in metres from origin. float
cam_z_pos World space z position of camera in metres from origin. float
cam_x_rot Rotation of camera around x axis in degrees. float
cam_y_rot Rotation of camera around y axis in degrees. float
cam_z_rot Rotation of camera around z axis in degrees. float
cam_FOV Camera field of view in degrees. For angles wider than 100 degrees a special purpose spherical camera should be used. float
cam_height Height of camera from ground in meters. float
cam_pitch Camera pitch in degrees -180 to +180. Positive is down. Zero is straight 
ahead.
float
cam_roll Camera roll in degrees -180 to +180. Positive is clockwise. No roll (flat) 
is zero.
float
cam_yaw Camera yaw in degrees -180 to +180. Positive is to the right. Zero is 
straight ahead.
float

File and image format data

Name Description Type
img_width Image file width in pixels. integer
img_height Image file height in pixels. integer
img_filename Name of image file that contains this object. string
mask_filename Name of object mask file. string
distance_filename Name of object distance file (i.e. pixel accurate radial distance 
from camera.
string
segment_filename Name of object segmented image file. string
normal_filename Name of object surface normals file. string

Camera relative velocity

Name Description Type
vel_x Relative to camera velocity. X component. Positive is to right of 
camera.
float
vel_y Relative to camera velocity. Y component. Positive is down. float
vel_z Relative to camera velocity. Z component. Positive is away from 
camera.
float

Position and facing data

Name Description Type
pos_x World space position x component. float
pos_y World space position y component float
pos_z World space position z component float
face_x World space facing direction x component. float
face_y World space facing direction y component float
face_z World space facing direction z component float
eye_left_pos_x World space position of the left eye x component float
eye_left_pos_y World space position of the left eye y component float
eye_left_pos_z World space position of the left eye z component float
eye_left_x World space left eye facing direction x component float
eye_left_y World space left eye facing direction y component float
eye_left_z World space left eye facing direction z component float
eye_right_pos_x World space position of the right eye x component float
eye_right_pos_y World space position of the right eye y component float
eye_right_pos_z World space position of the right eye z component float
eye_right_x World space right eye facing direction x component float
eye_right_y World space right eye facing direction y component float
eye_right_z World space right eye facing direction z component float

2D and 3D bounding boxes

Name Description Type
box_x1 1st X (left of 2nd X) in screen resolution coordinates. 0 is at left. float
box_y1 1st Y (above 2nd Y) in screen resolution coordinates. 0 is at top. float
box_x2 2nd X (further to right than 1st X) in screen resolution coordinates. 
Max X is at right.
float
box_y2 further down than 1st Y) in screen resolution coordinates. Max Y 
is bottom of screen.
float
cube_height Height of object when object is upright from bottom to top (meters). float
cube_width Width side to side of object when object is upright (meters). float
cube_length Length from front to back of object (meters). float
cube_dist_x Distance from camera center to object center (meters). X component 
of camera forward. Positive is to the right. Zero at center.
float
cube_dist_y Distance from camera centre to object centre (meters). Y component 
of camera forward. Positive is down. Zero at centre.
float
cube_dist_z Distance from camera center to object center (meters). Z component 
of camera forward. Positive is forward. Zero at camera.
float
cube_dist Distance from camera center to object center (meters). float
cube_alpha Degrees -180 to +180. Pose angle relative to camera forward. 
Deflection. Positive is to the right.
float
cube_beta Degrees -180 to +180. Pose angle relative to camera forward. Roll. 
Positive is clockwise.
float
cube_gamma Degrees -180 to +180. Pose angle relative to camera forward. 
Inclination. Positive is down.
float

Miscellaneous annotations

Name Description Type
tags String containing comma separated list of tags. Note that this string 
must be enclosed in double quotes.
string (JSON)
augmented Denotes whether the img_filename was generated by data 
augmentation. Default is 0 (ZERO) = NOT augmented.
integer
subset Denotes to which part of a dataset this object belongs. 0 for training, 
1 for validation, 2 for test. Default is -1 (MINUS ONE) = unassigned to 
any subset.
integer
used Denotes whether an object is being used for a dataset. Default is 1 
(ONE) = object will be used. This field is useful to mask out objects that 
will not be used during particular tasks.
integer
seg_color Segmentation colour of the object. string
seed object Seed, valid only when some randomiser has been used to 
generate the object such as for config actors
integer
source Source of data. For Highwai Simulator data, it's 'Highwai X.X' 
where X.X if the HAFS spec version.
string
notes Human readable notes field for each entry. Not used by software. string
time_code Absolute real-world time (UMT) as a string, YYYY-MM-DD-HH-MM-SS-NNNNNNNN, N=fraction of second. string