Leveraging Diffusion Perturbations for Measuring Fairness in Computer Vision

Authors: Nicholas Lui, Bryan Chia, William Berrios, Candace Ross, Douwe Kiela

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Using this dataset, we benchmark several vision-language models on a multiclass occupation classification task. We find that images generated with non-Caucasian labels have a significantly higher occupation misclassification rate than images generated with Caucasian labels, and that several misclassifications are suggestive of racial biases.
Researcher Affiliation Collaboration 1Stanford University 2Contextual AI 3Meta AI
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No We plan to release our code under a permissive license at this link: github.com/niclui/diffusion-perturbations.
Open Datasets Yes To enable greater exploration of our work, we release our gen-erated dataset at this link: bit.ly/occupation-dataset.
Dataset Splits No The paper does not specify explicit training/validation/test splits for the generated dataset, as its primary experiment is evaluating pre-trained models rather than training new ones on its own dataset.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions models like Stable Diffusion, Vi LT-B/32 VQA, CLIP, and FLAVA but does not provide specific software dependencies with version numbers (e.g., Python, PyTorch versions).
Experiment Setup No The paper states 'We discuss our choice of hyperparameters in Appendix A1,' indicating that specific experimental setup details like hyperparameters are not included in the main text.