ERENet: multi-level information fusion for refined and enhanced edge detection

Edge detection is a fundamental task in computer vision, pivotal for applications such as image segmentation, reconstruction, and object detection. Despite advancements, existing methods often struggle with accurately preserving fine-grained and low-contrast edges. This paper introduces ERENet, a novel edge refinement and enhancement network leveraging multi-level information fusion. ERENet employs a two-stage architecture featuring a feature extraction module and two plug-and-play modules: an edge enhancement module and an attention fusion module. These components synergistically refine edge lines and emphasize low-level texture details. Evaluated on the BIPED, BSDS500, UDED and NYUDv2 datasets, ERENet demonstrates superior performance across ODS, OIS, and AP metrics, providing clearer boundary features compared to state-of-the-art methods. Our code is available at https://github.com/shuang2099/ERENet.