GABDataset Experiments on Gender Bias in Vision-Language Models

This repository accompanies the research on the Gender-Activity Binding (GAB) bias in Vision-Language Models (VLMs). The GAB bias refers to the tendency of VLMs to incorrectly associate certain activities with a gender based on ingrained stereotypes. This research introduces the GAB dataset, comprising approximately 5,500 AI-generated images depicting various activities performed by individuals of different genders. The dataset is designed to assess and quantify the extent of gender bias in VLMs, particularly in text-to-image and image-to-text retrieval tasks.

Our experiments reveal that VLMs experience a significant drop in performance when the gender of the person performing an activity does not align with stereotypical expectations. Specifically, the presence of an unexpected gender performing a stereotyped activity leads to an average performance decline of about 13.2% in image-to-text retrieval tasks. Additionally, when both genders are present in the scene, the models are often biased toward associating the activity with the gender expected to perform it. The study also explores the bias in text encoders and their role in the gender-activity binding phenomenon.

Below is an overview of the creation process of the GAB dataset and the empirical tests conducted to assess the gender-activity binding bias:

In this repository, we provide the code and dataset (GABDataset) used to examine gender bias in Vision-Language Models (VLMs) through various experiments described in the main paper. The repository is organized into three phases: phaze1, phaze2, and phaze3, with corresponding directories for each experiment.

Repository Structure

1. BiasExperiment Directory

This directory contains the notebooks for measuring subject binding bias as detailed in the paper. For each phase (phaze1, phaze2, phaze3), you will find:

• Experiment Notebook: Used to conduct the experiment.
• Results Notebook: Used to aggregate and plot the results.

2. TextEncoderBiasExperiment Directory

This directory focuses on measuring text encoder bias. It includes:

• Experiment Notebook: Used to perform the experiment.
• Results Notebook: Used to aggregate the results.

3. TextToImageRetrievalExperiment Directory

In this directory, we assess the model's capability for text-to-image retrieval. It includes:

• Experiment Notebook: Used to execute the experiment.
• Results Notebook: Used to aggregate and plot the results.

4. ActivityRetrievalExperiment Directory

This directory measures the model’s ability to differentiate and understand activities in a scene between similar actions. For each phase, the directory contains:

• Experiment Notebook: Used to conduct the experiment.
• Results Notebook: Used to aggregate and plot the results.

Phases

The repository is structured into three phases:

• phaze1
• phaze2
• phaze3

Each phase represents a distinct set of images used in the experiments.

How to Use

1. Clone this repository to your local machine.
2. Download and unzip the dataset by running the following commands:

   cd GABInsight
   gdown https://drive.google.com/uc?id=13qeOuszF52b8F7Bkvxg5GEEHEl_7obzM
   unzip phazes.zip

3. Navigate to the directory of the experiment you wish to explore.
4. Open the corresponding experiment notebook to run the experiment.
5. Use the results notebook to aggregate and visualize the results.

For detailed information on each experiment, refer to the main paper associated with this repository.