Discovering Object Masks with Transformers for Unsupervised Semantic Segmentation

TL;DR

MaskDistill introduces a data-driven, unsupervised framework for semantic segmentation that generates and refines object masks without handcrafted priors, outperforming previous methods on PASCAL and COCO datasets.

Contribution

The paper proposes MaskDistill, a novel unsupervised segmentation framework that eliminates handcrafted priors and uses clustering and filtering of object masks for improved accuracy.

Findings

Outperforms previous methods on PASCAL (+11% mIoU) and COCO (+4% mask AP50)

Does not rely on low-level image cues or object-centric assumptions

Generates high-quality object masks without handcrafted priors

Abstract

The task of unsupervised semantic segmentation aims to cluster pixels into semantically meaningful groups. Specifically, pixels assigned to the same cluster should share high-level semantic properties like their object or part category. This paper presents MaskDistill: a novel framework for unsupervised semantic segmentation based on three key ideas. First, we advocate a data-driven strategy to generate object masks that serve as a pixel grouping prior for semantic segmentation. This approach omits handcrafted priors, which are often designed for specific scene compositions and limit the applicability of competing frameworks. Second, MaskDistill clusters the object masks to obtain pseudo-ground-truth for training an initial object segmentation model. Third, we leverage this model to filter out low-quality object masks. This strategy mitigates the noise in our pixel grouping prior and results in a clean collection of masks which we use to train a final segmentation model. By combining these components, we can considerably outperform previous works for unsupervised semantic segmentation on PASCAL (+11% mIoU) and COCO (+4% mask AP50). Interestingly, as opposed to existing approaches, our framework does not latch onto low-level image cues and is not limited to object-centric datasets. The code and models will be made available.