APS-NeuS: Adaptive planar and skip-sampling for 3D object surface reconstruction in high-specular scenes

Abstract

High-fidelity 3D object surface reconstruction remains challenging in real-world scenes with strong specular reflections, where multi-view consistency is disrupted by reflection artifacts. To address this, we propose APS-NeuS, an implicit neural rendering framework designed to robustly separate target objects from reflective interference. Specifically, we establish a pixel-wise auxiliary mirror plane to differentiate reflections from target objects and incorporate a Laplacian gradient to better recover their edges and fine structures. Additionally, we introduce a skip-sampling strategy to reduce the impact of reflective interference, further enhancing multi-view consistency and surface fidelity. Finally, we introduce an exclusion loss to help the model more accurately separate the target objects from the reflective parts during initialization by comparing the gradient differences. Extensive experiments on synthetic and real-world datasets show that APS-NeuS achieves superior reconstruction quality under high-specular reflection conditions, demonstrating its practical applicability to complex environments. Code is available at https://github.com/ujsjl/APS-NeuS.