🤖 AI Summary
To address inaccurate object localization and blurry detail reconstruction in high-resolution binary image segmentation (DIS), this paper proposes BiRefNet, a bilateral reference network. Methodologically, BiRefNet adopts a two-stage localization-reconstruction framework: the localization module fuses global semantic cues for coarse-grained object localization, while the reconstruction module introduces a novel bilateral reference (BiRef) mechanism—jointly leveraging hierarchical source-side image patches and target-side gradient maps for fine-grained reconstruction, enhanced by gradient-aware supervision to strengthen edge modeling. Furthermore, we design a DIS-specific training strategy. Extensive experiments demonstrate that BiRefNet achieves state-of-the-art performance across four mainstream DIS benchmarks, significantly improving both segmentation accuracy and boundary sharpness. The code is publicly available.
📝 Abstract
We introduce a novel bilateral reference framework (BiRefNet) for high-resolution dichotomous image segmentation (DIS). It comprises two essential components: the localization module (LM) and the reconstruction module (RM) with our proposed bilateral reference (BiRef). The LM aids in object localization using global semantic information. Within the RM, we utilize BiRef for the reconstruction process, where hierarchical patches of images provide the source reference and gradient maps serve as the target reference. These components collaborate to generate the final predicted maps. We also introduce auxiliary gradient supervision to enhance focus on regions with finer details. Furthermore, we outline practical training strategies tailored for DIS to improve map quality and training process. To validate the general applicability of our approach, we conduct extensive experiments on four tasks to evince that BiRefNet exhibits remarkable performance, outperforming task-specific cutting-edge methods across all benchmarks. Our codes are available at https://github.com/ZhengPeng7/BiRefNet.