BEN: Using Confidence-Guided Matting for Dichotomous Image Segmentation
Maxwell Meyer, Jack Spruyt
TL;DR
The paper addresses the challenge of precise foreground segmentation in dichotomous image segmentation (DIS) by uniting matting and segmentation through Confidence-Guided Matting (CGM). It introduces Background Erase Network (BEN) with a two-stage pipeline: BEN Base provides initial segmentation and BEN Refiner, guided by confidence-based trimaps, refines uncertain regions to enhance boundary accuracy. On the DIS5K validation set (DIS-VD), BEN achieves state-of-the-art results, with the BEN Base+Refiner outperforming strong baselines such as MVANet and DiffDIS by substantially improving edge quality and overall metrics. This work establishes a new paradigm for integrating matting techniques into high-precision segmentation and suggests extending confidence-based matting to broader segmentation tasks in future work.
Abstract
Current approaches to dichotomous image segmentation (DIS) treat image matting and object segmentation as fundamentally different tasks. As improvements in image segmentation become increasingly challenging to achieve, combining image matting and grayscale segmentation techniques offers promising new directions for architectural innovation. Inspired by the possibility of aligning these two model tasks, we propose a new architectural approach for DIS called Confidence-Guided Matting (CGM). We created the first CGM model called Background Erase Network (BEN). BEN consists of two components: BEN Base for initial segmentation and BEN Refiner for confidence-based refinement. Our approach achieves substantial improvements over current state-of-the-art methods on the DIS5K validation dataset, demonstrating that matting-based refinement can significantly enhance segmentation quality. This work introduces a new paradigm for integrating matting and segmentation techniques, improving fine-grained object boundary prediction in computer vision.