Towards Agnostic and Holistic Universal Image Segmentation with Bit Diffusion

TL;DR

This paper proposes a diffusion-based universal image segmentation framework that operates agnostically without masks, introduces key adaptations for discrete data, and offers new capabilities like ambiguity modeling, aiming to close the performance gap with mask-based methods.

Contribution

It introduces a novel diffusion model framework for universal segmentation, with key adaptations for discrete data and capabilities like ambiguity modeling, advancing beyond traditional mask-based approaches.

Findings

Location-aware palette improves performance

Sigmoid loss weighting outperforms alternatives

Model narrows gap with mask-based architectures

Abstract

This paper introduces a diffusion-based framework for universal image segmentation, making agnostic segmentation possible without depending on mask-based frameworks and instead predicting the full segmentation in a holistic manner. We present several key adaptations to diffusion models, which are important in this discrete setting. Notably, we show that a location-aware palette with our 2D gray code ordering improves performance. Adding a final tanh activation function is crucial for discrete data. On optimizing diffusion parameters, the sigmoid loss weighting consistently outperforms alternatives, regardless of the prediction type used, and we settle on x-prediction. While our current model does not yet surpass leading mask-based architectures, it narrows the performance gap and introduces unique capabilities, such as principled ambiguity modeling, that these models lack. All models were trained from scratch, and we believe that combining our proposed improvements with large-scale pretraining or promptable conditioning could lead to competitive models.