Unbiased Image Synthesis via Manifold Guidance in Diffusion Models
This addresses bias issues in generative models for skewed datasets like CelebA, offering an unsupervised solution to improve diversity, though it is incremental as it builds on existing diffusion models.
The paper tackles the problem of bias in diffusion models, where generated images inadvertently favor certain attributes, such as gender in CelebA, with female representation amplified by 148% compared to the dataset's 57.9% bias. The result is a plug-and-play method that mitigates this bias without model modifications, enhancing image quality and unbiasedness.
Diffusion Models are a potent class of generative models capable of producing high-quality images. However, they often inadvertently favor certain data attributes, undermining the diversity of generated images. This issue is starkly apparent in skewed datasets like CelebA, where the initial dataset disproportionately favors females over males by 57.9%, this bias amplified in generated data where female representation outstrips males by 148%. In response, we propose a plug-and-play method named Manifold Guidance Sampling, which is also the first unsupervised method to mitigate bias issue in DDPMs. Leveraging the inherent structure of the data manifold, this method steers the sampling process towards a more uniform distribution, effectively dispersing the clustering of biased data. Without the need for modifying the existing model or additional training, it significantly mitigates data bias and enhances the quality and unbiasedness of the generated images.