Stochastic Segmentation with Conditional Categorical Diffusion Models
This addresses the need for stochastic semantic segmentation in domains like medical diagnostics and autonomous driving, where multiple valid annotations exist, representing an incremental improvement over existing methods.
The paper tackled the problem of generating multiple possible correct segmentation maps to reflect aleatoric uncertainty in safety-critical domains, proposing a conditional categorical diffusion model that achieved state-of-the-art performance on LIDC and outperformed baselines on Cityscapes.
Semantic segmentation has made significant progress in recent years thanks to deep neural networks, but the common objective of generating a single segmentation output that accurately matches the image's content may not be suitable for safety-critical domains such as medical diagnostics and autonomous driving. Instead, multiple possible correct segmentation maps may be required to reflect the true distribution of annotation maps. In this context, stochastic semantic segmentation methods must learn to predict conditional distributions of labels given the image, but this is challenging due to the typically multimodal distributions, high-dimensional output spaces, and limited annotation data. To address these challenges, we propose a conditional categorical diffusion model (CCDM) for semantic segmentation based on Denoising Diffusion Probabilistic Models. Our model is conditioned to the input image, enabling it to generate multiple segmentation label maps that account for the aleatoric uncertainty arising from divergent ground truth annotations. Our experimental results show that CCDM achieves state-of-the-art performance on LIDC, a stochastic semantic segmentation dataset, and outperforms established baselines on the classical segmentation dataset Cityscapes.