DiffInject: Revisiting Debias via Synthetic Data Generation using Diffusion-based Style Injection

Donggeun Ko, Sangwoo Jo, Dongjun Lee, Namjun Park, Jaekwang KIM

Keywords: Debiasing, Synthetic Data, Diffusion Model

TL;DR: Using our diffusion-based framework, DiffInject, we inject bias-conflict features into bias-aligned samples to generate synthetic data, which is subsequently employed to debias the biased classifier.

Abstract: Dataset bias is a significant challenge in machine learning, where specific attributes, such as texture or color of the images are unintentionally learned resulting in detrimental performance. To address this, previous efforts have focused on debiasing models either by developing novel debiasing algorithms or by generating synthetic data to mitigate the prevalent dataset biases. However, generative approaches to date have largely relied on using bias-specific samples from the dataset, which are typically too scarce. In this work, we propose, DiffInject, a straightforward yet powerful method to augment synthetic bias-conflict samples using a pretrained diffusion model. This approach significantly advances the use of diffusion models for debiasing purposes by manipulating the latent space. Our framework does not require any explicit knolwedge of the bias types or labelling, making it a fully unsupervised setting for debiasing. Our methodology demonstrates substantial result in effectively reducing dataset bias.

Submission Number: 27