FACET; Fairness in Computer Vision Evaluation Benchmark

FACET: Fairness in Computer Vision Evaluation Benchmark

Gustafson, Laura, et al. "Facet: Fairness in computer vision evaluation benchmark." Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023.

참고:

• https://aipapersacademy.com/facet/
• https://arxiv.org/pdf/2309.00035

Contents

1. Introduction
2. Examples
3. Statistics

1. Introduction

FACET

( = Fairness in Computer Vision Evaluation Benchmark )

• Dataset to evaluate a benchmark for fairness of computer vision models
• Previous works: have biases (example below)

2. Examples

Few more annotations, other than class

• e.g., perceived gender presentation, skin tone …

3. Statistics

Details

• Contains 32k images and 50k people ( including images with more than 1 people )

• Each person is also surrounded with a bounding box

  \(\rightarrow\) Can be used for object detection

• 69k masks labeled as person, hair or clothing

  \(\rightarrow\) Can be used for image segmentation

• Various more attributes

  • e.g., hair details, face masks, tattoo and lighting..