OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes
Sepehr Dehdashtian, Gautam Sreekumar, Vishnu Naresh Boddeti
TL;DR
This paper introduces OASIS, a sociologically aligned quantitative measure for stereotypes in text-to-image models, revealing persistent biases in state-of-the-art models despite high image quality.
Contribution
The paper proposes OASIS, a novel framework with two scores and two methods to quantify and analyze stereotypes in T2I models, aligning with sociological definitions.
Findings
Newer T2I models still exhibit strong stereotypes.
Stereotype prevalence increases for nationalities with lower Internet presence.
OASIS effectively uncovers origins and extent of stereotypes in generated images.
Abstract
Images generated by text-to-image (T2I) models often exhibit visual biases and stereotypes of concepts such as culture and profession. Existing quantitative measures of stereotypes are based on statistical parity that does not align with the sociological definition of stereotypes and, therefore, incorrectly categorizes biases as stereotypes. Instead of oversimplifying stereotypes as biases, we propose a quantitative measure of stereotypes that aligns with its sociological definition. We then propose OASIS to measure the stereotypes in a generated dataset and understand their origins within the T2I model. OASIS includes two scores to measure stereotypes from a generated image dataset: (M1) Stereotype Score to measure the distributional violation of stereotypical attributes, and (M2) WALS to measure spectral variance in the images along a stereotypical attribute. OASIS also includes two…
Peer Reviews
Decision·ICLR 2025 Spotlight
1. Measuring sterotyping is an important and challenging problem for generative models. 2. The diversity of metrics seem quite interesting to get a more complete picture of how bias and stereotypes arise in these models. 3. The paper is fairly thorough in its testing.
1. The biggest weakness in my opinion is that the paper is quite hard to follow for many of the metrics. I believe this is due to making the description of the metrics unnecessarily complex, and I would encourage the authors to try to describe all of the concepts more simply. In addition to making the paper harder to understand, it also obfuscates the insights and contributions. 2. One of the papers claimed main contributions is measuring stereotyping relative to societal priors on the distrib
Addressing a research gap Clear mythology Application to existing models
It would have been nice if the writeup included a detailed case study of qualitative analysis of cases of stereotypes
- The paper is well-written, clear, and easy to follow. - The paper addresses a complex and socially significant problem by evaluating biases and stereotypes in T2I models. Additionally, it provides new methods to trace the origins of these stereotypes in T2I models, offering valuable insights to help mitigate these biases. - The findings of this work highlight the need for greater inclusion and representation of underrepresented communities, particularly those from regions with limited internet
- This work focuses exclusively on generated images based on nationality, which raises several concerns. For instance, in racially diverse countries like the United States, how can an image indicate that someone is a U.S. citizen without relying on stereotypes and social biases? Additionally, in regions where people from neighboring countries may share similar facial features or cultures (such as North and South Korea), how useful can OASIS be in distinguishing these cases? - How can OASIS be us
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsIoT and Edge/Fog Computing · Big Data and Business Intelligence · Business Process Modeling and Analysis
MethodsOASIS · ALIGN