Table 1 Dataset comparison by intersectional subgroups

From: Fair human-centric image dataset for ethical AI benchmarking

Intersectional group

Dataset

Number of subgroups

Number of images

   

Med.

Max.

Min.

Gender × age

FHIBE

23

23

3,353

1

CCv1

12

220

523

1

CCv2

23

23

1,598

1

FACET

9

2,070

22,008

3

MIAP

4

7,439

21,195

254

Gender × age × skin tone

FHIBE

92

42

1,168

1

CCv1

62

38

129

1

CCv2

137

5

909

1

FACET

82

284

12,506

1

Gender × age × ancestry region

FHIBE

72

128

8,415

4

Gender × age × ancestry subregion

FHIBE

322

31

1,683

1

Gender × age × ancestry region × skin tone

FHIBE

275

36

5,645

4

Gender × age × ancestry subregion × skin tone

FHIBE

1,234

9

1,129

1

  1. This table shows how FHIBE compares with other fairness evaluation datasets based on intersectional groups, including gender or pronoun (only FHIBE uses pronouns), age, ancestry and skin tone. Subgroup counts and the median (med.), minimum (min.) and maximum (max.) number of images per subgroup are shown. FHIBE offers broader demographic representation through comprehensive annotations. Note that FACET and FHIBE images may be counted in multiple attribute categories if they have multiple/nested annotations (for example, multiple gender/pronoun or skin tone selections).
Back to article page