REVISE: REvealing VIsual biaSEs
A tool for automatically detecting and suggesting mitigation for object, attribute, and geography-based biases in visual datasets.
What is REVISE: REvealing VIsual biaSEs?
REVISE is a research tool for measuring and mitigating bias in visual datasets. It automatically detects potential biases along object-based, attribute-based, and geography-based patterns, providing actionable insights to improve dataset fairness. The tool helps identify imbalances in representation, attribute distribution, and geographic coverage that could lead to skewed computer vision models.
Computer vision researchers, data scientists, and AI ethics practitioners who need to audit visual datasets for fairness before model training. It is particularly useful for teams building or curating large-scale image datasets for machine learning.
REVISE offers a comprehensive, multi-axis approach to bias detection that goes beyond simple demographic checks. Its automated measurement pipelines and visual summaries enable systematic dataset auditing, reducing manual effort and providing clear pathways for mitigation.
Overview
REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasets --- https://arxiv.org/abs/2004.07999
Use Cases
Best For
- Auditing image datasets for representation biases before model training
Related Projects
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
