IARPA Janus  0.4.0
IARPA Janus Program API
Public Attributes | List of all members
janus_attributes Struct Reference

Attributes for a particular object in an image. More...

#include <iarpa_janus.h>

Public Attributes

double detection_confidence
 A higher value indicates greater detection confidence.
 
double face_x
 Horizontal offset to top-left corner of face (pixels) More...
 
double face_y
 Vertical offset to top-left corner of face (pixels) More...
 
double face_width
 Face horizontal size (pixels) More...
 
double face_height
 Face vertical size (pixels) More...
 
double right_eye_x
 Face landmark (pixels) More...
 
double right_eye_y
 Face landmark (pixels) More...
 
double left_eye_x
 Face landmark (pixels) More...
 
double left_eye_y
 Face landmark (pixels) More...
 
double nose_base_x
 Face landmark (pixels) More...
 
double nose_base_y
 Face landmark (pixels) More...
 
double face_yaw
 Face yaw estimation (degrees).
 
double forehead_visible
 Visibility of forehead. More...
 
double eyes_visible
 Visibility of eyes. More...
 
double nose_mouth_visible
 Visibility of nose and mouth. More...
 
double indoor
 Image was captured indoors. More...
 
double gender
 Gender of subject of interest, 1 for male, 0 for female.
 
double age
 Approximate age of subject (years) More...
 
double skin_tone
 Skin tone of subject. More...
 
double frame_rate
 Frames per second, or 0 for images.
 

Detailed Description

Attributes for a particular object in an image.

Attributes associated with an object. Values of NaN (isnan) indicate that the attribute value is unknown.

Attribute Definitions

Attributes collected for images in the Janus program were annotated using crowd sourcing. As a result, attributes are defined in layman's terms, instead of strict scientific definitions. Below are the instructions given to workers for each attribute, which should act as a de-facto definition of the attribute. In some cases instructions evolved slightly over time, reflecting lessons learned from communicating with the workers.

As implementations of the Janus API leverage divergent algorithms and data sets, the precise semantics of each attribute may differ by implementation.

Face

Instructions with images.

We strongly prefer tight bounding boxes over loose ones. There's a bit of leeway here, especially because it is also important to capture the whole head, but if the bounding boxes are too loose we will have to reject the work. For the case of people wearing hats or headgear, or who just have large hair, try to approximate where the head would be (it's alright if you're not exact). As a general rule of thumb, if you can see a feature on a face (an eye, nose, or mouth) or if you can tell it's a forward facing head (but it's too far away to make out individual features), box it! Please err on the side of boxing heads, the exceptions being heads that are completely facing away and heads that are blocked so you cannot see any features. It's okay if boxes overlap! But, try to only box the parts of the head you can see. Non-living faces (portraits, faces on screens, etc.) are fair game, too! If they look realistic, box them. We understand that some images aren't worth your time. In the case of image shown below, where there are many small, blurry heads in the background, just box the person in focus (or press "Not visible" if no one is in focus). In cases like this one, where many faces are distinguishable, please press the "Crowd" button. We ask that you only do this in cases where there truly are crowds - at least 25 heads or so. These images will definitely be reviewed, and wrongly pressing this button will result in a rejection.

Right Eye

For this task, click on the middle of the right eye of the subject shown. Please note that the right eye of the subject will usually appear on the left. Also note that the middle of the eye is not necessarily the pupil, but rather the midpoint between outer edges of the eye. If the right eye is not visible or too blurry to identify, press "Not Visible" - unless the eye is obscured by dark glasses, in which case we ask that you estimate where the eye would be.

Left Eye

For this task, click on the middle of the left eye of the subject shown. Please note that the left eye of the subject will usually appear on the right. Also note that the middle of the eye is not necessarily the pupil, but rather the midpoint between outer edges of the eye. If the left eye is not visible or too blurry to identify, press "Not Visible" - unless the eye is obscured by dark glasses, in which case we ask that you estimate where the eye would be.

Nose Base

For this task, click on the middle of the nose base of the subject shown. This is considered to be the middle of where the nose meets the face. If the nose base is not visible or too blurry to identify, press "Not Visibleā€.

Forehead Occlusion

For this task, choose "Forehead Covered" when:

Eye Occlusion

For this task, choose "Eyes Covered" when:

Nose and Mouth Occlusion

For this task, choose "Covered" when:

Indoor or Outdoor

For this task, choose "Indoor" when:

Age

For this task, estimate the approximate age of the person shown. The age categories in years and corresponding values in the janus_metadata , are:

0 - 19 20 - 34 35 - 49 50-64 65+ Unknown
1 2 3 4 5 0

If image quality is low, please take your best guess.

Skin Tone

Skin tone is generalized into 6 categories described below, along with their corresponding values in the janus_metadata.

Light Pink Light Yellow Medium Pink/Brown Medium Yellow/Brown Medium-Dark Brown Dark Brown
1 2 3 4 5 6

Facial Hair

For this task, there are four possible types of facial hair. Select the facial hair type that is closest to the description below.

No Facial Hair Moustache Goatee Beard
0 1 2 3

Member Data Documentation

double janus_attributes::age

Approximate age of subject (years)

See also
Age.
double janus_attributes::eyes_visible

Visibility of eyes.

See also
Eye Occlusion.
double janus_attributes::face_height

Face vertical size (pixels)

See also
Face.
double janus_attributes::face_width

Face horizontal size (pixels)

See also
Face.
double janus_attributes::face_x

Horizontal offset to top-left corner of face (pixels)

See also
Face.
double janus_attributes::face_y

Vertical offset to top-left corner of face (pixels)

See also
Face.
double janus_attributes::forehead_visible

Visibility of forehead.

See also
forehead_visible.
double janus_attributes::indoor

Image was captured indoors.

See also
Indoor or Outdoor.
double janus_attributes::left_eye_x

Face landmark (pixels)

See also
Left Eye.
double janus_attributes::left_eye_y

Face landmark (pixels)

See also
Left Eye.
double janus_attributes::nose_base_x

Face landmark (pixels)

See also
Nose Base.
double janus_attributes::nose_base_y

Face landmark (pixels)

See also
Nose Base.
double janus_attributes::nose_mouth_visible

Visibility of nose and mouth.

See also
nouse_mouth_visible.
double janus_attributes::right_eye_x

Face landmark (pixels)

See also
Right Eye.
double janus_attributes::right_eye_y

Face landmark (pixels)

See also
Right Eye.
double janus_attributes::skin_tone

Skin tone of subject.

See also
Skin Tone.

The documentation for this struct was generated from the following file: